BEGIN:VCALENDAR
VERSION:2.0
METHOD:PUBLISH
PRODID:-//Example Calendar//EN
CALSCALE:GREGORIAN
X-WR-CALNAME:Technion - Computer Science Faculty Calendar
BEGIN:VTIMEZONE
TZID:Asia/Jerusalem
BEGIN:STANDARD
DTSTART:19500910T020000
TZOFFSETFROM:+0300
TZOFFSETTO:+0200
RRULE:FREQ=YEARLY;BYMINUTE=0;BYHOUR=2;BYDAY=-1SU;BYMONTH=10
END:STANDARD
BEGIN:DAYLIGHT
DTSTART:19500331T020000
TZOFFSETFROM:+0200
TZOFFSETTO:+0300
RRULE:FREQ=YEARLY;BYMINUTE=0;BYHOUR=2;BYDAY=-1FR;BYMONTH=3
END:DAYLIGHT
END:VTIMEZONE
X-WR-TIMEZONE:Asia/Jerusalem
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20230102T173000
DTEND;TZID=Asia/Jerusalem:20230102T173000
DTSTAMP;TZID=Asia/Jerusalem:20230102T173000
SUMMARY: CSpecial Event   Recruitment Day by Microsoft  at 2023-01-02 17:30:00
DESCRIPTION:You are invited to Microsoft Recruitment day, to a question and answer session with the team about the company's recruitment process, and to a lecture by Noa Berman, a software developer at Microsoft Security, on: Ransomware attacks and how we defend against them at Microsoft Defender for Endpoint, on Monday, January 2, 17:30, at Taub 9.\n \nPlease pre-register.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 9
UID:eventx6a5a287eeacf710340
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20230103T103000
DTEND;TZID=Asia/Jerusalem:20230103T123000
DTSTAMP;TZID=Asia/Jerusalem:20230103T103000
SUMMARY: CSpecial Event   Yahoo Research at CS  at 2023-01-03 10:30:00
DESCRIPTION:Yahoo Research will visit CS for a special meeting with graduate students on Tuesday, January 3, 2023 starting at 10:30 at the Grads Club, 2nd floor (at the end of the corridor), Taub Computer Science Building: \n\nProgram: \n10:30 - Gathering \n11:00 - Intro - Yahoo Israel Research Center \n11:15 - Lecture 1: Leveraging User Email Actions to Improve Ad-Close Prediction - by Yaroslav Fyodorov \n11:35 - Lecture 2: Augmentation for Consistent Categorization - by Stav Yanovsky Daye \n11:55 - Lecture 3: Consistent Text Categorization with Data Augmentation - by Alex Shtoff \n12:00 - Mingling \n\nPlease register in advance
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Grads Club, CS Taub 2nd floor
UID:eventx6a5a287eead1e10348
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20230103T110000
DTEND;TZID=Asia/Jerusalem:20230103T120000
DTSTAMP;TZID=Asia/Jerusalem:20230103T110000
SUMMARY: colloq  talk by Brit Youngmann (CSAIL MIT)  CS Lecture: Data Tools for Accelerated Scientific Discoveries  at 2023-01-03 11:00:00
DESCRIPTION:Causal inference is fundamental to empirical research in natural and social sciences and is essential for scientific discoveries. Two key challenges for conducting causal inference are (i) acquiring all attributes required for the analysis, and (ii) identifying which attributes should be included in the analysis. Failing to include all necessary attributes may lead to false discoveries and erroneous conclusions. However, in real-world settings, analysts may only have access to partial data. Further, to identify which attributes should be included in the analysis, analysts critically rely on domain knowledge, often given in the form of a causal DAG. However, such domain knowledge is often unavailable and cannot be fully recovered from data. In this talk we will present two works that address these challenges by leveraging data management techniques and ideas. \n\nBio:\nBrit is a postdoc researcher at CSAIL MIT, working with Prof. Michael Cafarella. She received her Ph.D. at Tel-Aviv University under the supervision of Prof. Tova Milo. Her research is centered around informative and responsible data science and causal analysis. Brit is the recipient of several awards, including the data science fellowship for outstanding Ph.D. students of the planning and budgeting committee of the Israeli council for higher education (VATAT), the Schmidt postdoctoral award for women in mathematical and computing sciences, and the planning and budgeting committee of the Israeli council for higher education (VATAT) postdoctoral scholarship in Data Science. She served on multiple program committees, including at the SIGMOD and ICDE conferences.  \n
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 601
UID:eventx6a5a287eead3610347
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20230103T143000
DTEND;TZID=Asia/Jerusalem:20230103T153000
DTSTAMP;TZID=Asia/Jerusalem:20230103T143000
SUMMARY: colloq  talk by Michael Lustig (UC Berkeley)  CS Colloquia: Adventures in Computational MRI  at 2023-01-03 14:30:00
DESCRIPTION:Magnetic resonance imaging (MRI) is a powerful, ionizing-radiation-free medical imaging modality. The vast physical and physiological parameters, which MRI is sensitive to, makes it possible to visualize both structure and function in the body. However the prolonged time necessary to capture the information in this large parameter space remains a major limitation of this phenomenal modality, which the field of computational MRI aims to address. By computational MRI we refer to the joint optimization of the imaging system hardware, the data encoding, the data acquisition  and  the image reconstruction together. In this talk I will describe some of the efforts my group has been engaged in towards mitigating with motion and dynamics that occurs during MRI scanning, in particular when performing body imaging of pediatric patients. Specifically I will focus on unsupervised and supervised methods for dynamic 2D and 3D imaging and  learning based high fidelity reconstructions of fine structures and textures.\n \nShort bio:\nMichael (Miki) Lustig is a Professor in Electrical Engineering and Computer Science.  He joined the faculty of the EECS Department at UC Berkeley in Spring 2010. He received his B.Sc. in Electrical Engineering from the Technion, Israel Institute of Technology in 2002. He received his MSc and Ph.D. in Electrical Engineering from Stanford University in 2004 and 2008, respectively. His research focuses on computational imaging methods in medical imaging, particularly Magnetic Resonance Imaging (MRI)— these include a spectrum of work ranging from Hardware, through MRI pulse sequences and acquisitions, Image reconstruction and clinical applications of MRI. Miki is a Fellow of the Society of Magnetic Resonance in Medicine.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Room 337 taub bld.
UID:eventx6a5a287eead4a10336
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20230104T103000
DTEND;TZID=Asia/Jerusalem:20230104T113000
DTSTAMP;TZID=Asia/Jerusalem:20230104T103000
SUMMARY: colloq  talk by Yoav Levine (AI21 Labs)  CS Lecture: Theoretical and practical principles for designing, training, and deploying huge language models  at 2023-01-04 10:30:00
DESCRIPTION:The field of natural language processing (NLP) has been advancing in giant strides over the past several years. The main drivers of this success are: (1) scaling the Transformer deep network architecture to unprecedented sizes and (2) “pretraining” the Transformer over massive amounts of unlabeled text. In this talk, I will describe efforts to provide principled guidance for the above main components and further thrusts in contemporary NLP, aimed to serve as timely constructive feedback for the strong empirical pull in this field.\n \nI will begin by describing our theoretical framework for analyzing Transformers, and present results on the depth to width tradeoffs in Transformers, identified bottlenecks within internal Transformer dimensions, and identified biases introduced during the Transformer self-supervised pretraining phase. This framework has guided the design and scale of several of the largest existing language models, including Chinchilla by Deepmind (70 billion learned parameters), Bloom by BigScience (176 billion learned parameters), and Jurassic-1 by AI21(178 billion learned parameters). Then, I will describe our works on leveraging linguistic biases such as word senses or frequent n-grams in order to increase efficiency of the self-supervised pretraining phase. Subsequently, I will describe novel principles for addressing a present-day problem stemming from the above success of scaling, namely, how to deploy a huge language model such that it specializes in many different use cases simultaneously (e.g., supporting many different customer needs simultaneously). Finally, I will comment on future challenges in this field, and will relatedly present a recent theoretical result on the importance of intermediate supervision when solving composite NLP tasks.\n \nThis talk is based on works published in NeurIPS 2020, ACL 2020, ICLR 2021 (spotlight), ICML 2021, ICLR 2022 (spotlight), ICML 2022 (workshop), as well as on several recent preprints.\n\nBio:\nDr. Yoav Levine serves as co-Chief Scientist at AI21 Labs, an Israeli start up in the field of NLP. He earned his PhD at the Hebrew University, under the supervision of Prof. Amnon Shashua. His PhD studies were supported by the Israeli Academy of Sciences Adams fellowship, and for them he has received the Blavatnik PhD Prize awarded to the top 5 Israeli PhD theses in the field of computer science. Prior to his doctoral studies, he earned an M.Sc. in theoretical condensed matter physics from the Weizmann Institute of Science under the supervision of Prof. Yuval Oreg, and a double B.Sc. in physics and electrical engineering (both summa cum laude) from Tel Aviv University, supported by the Adi Lautman excellence program.\n\n
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 601
UID:eventx6a5a287eead5c10346
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20230104T123000
DTEND;TZID=Asia/Jerusalem:20230104T123000
DTSTAMP;TZID=Asia/Jerusalem:20230104T123000
SUMMARY: CSpecial Event   Recruitment Day and Workshop by Intel  at 2023-01-04 12:30:00
DESCRIPTION:Intel will hold a recruitment day and will present employment opportunities, as well as a "Fusion 360" workshop of 3D modeling, printing and Makers experience, on Wednesday, January 4, 2023, starting at 12:30 in the Taub lobby.\n\nFor the workshop please pre-register in advance (the number of places for the workshop is limited and subject to confirmation of registration)
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:CS Taub Lobby
UID:eventx6a5a287eead7210342
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20230104T123000
DTEND;TZID=Asia/Jerusalem:20230104T133000
DTSTAMP;TZID=Asia/Jerusalem:20230104T123000
SUMMARY: Theory Seminar  talk by Ilya Volkovich (Boston College)  Theory Seminar: Mutual Empowerment between Circuit Obfuscation and Circuit Minimization  at 2023-01-04 12:30:00
DESCRIPTION:We study close connections between Indistinguishability Obfuscation (IO) and the Minimum Circuit Size Problem (MCSP), and argue that algorithms for one of MCSP or IO would empower the other one. Some of our main results are:\n\nIf there exists a perfect (imperfect) IO that is computationally-secure against non-uniform polynomial-size circuits, then we obtain fixed-polynomial lower bounds against NP(MA).\n\nIn addition, computationally-secure IO against non-uniform polynomial-size circuits imply super-polynomial lower bounds against NEXP.\n\nIf MCSP is in BPP, then statistical security and computational security for IO are equivalent.\n\nTo the best of our knowledge, this is the first consequence of strong circuit lower bounds from the existence of an IO. The results are obtained via a construction of an optimal universal distinguisher, computable in randomized polynomial time with access to the MCSP oracle, that will distinguish any two circuit-samplable distributions with the advantage that is the statistical distance between these two distributions minus some negligible error term. This is our main technical contribution. As another application, we get a simple proof of the result by Allender and Das (Inf. Comput., 2017) that SZK is contained in BPP^MCSP.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 201
UID:eventx6a5a287eead8210344
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20230105T143000
DTEND;TZID=Asia/Jerusalem:20230105T153000
DTSTAMP;TZID=Asia/Jerusalem:20230105T143000
SUMMARY: MSC  talk by Amit Ganz  Online Submodular Welfare Maximization with General Utilities  at 2023-01-05 14:30:00
DESCRIPTION:We consider the online Submodular Welfare problem.\n\nIn this problem we are given n bidders each equipped with a submodular utility and m items that arrive online. \n\nThe goal is to assign each item, once it arrives, to a bidder or discard it, while maximizing the sum of utilities.\n\nThe case of monotone utilities has attracted much attention, however much less is known once utilities are general and not necessarily monotone.\n\nWhen an adversary determines the items' arrival order, we present an algorithm, inspired by the algorithm of [Dobzinski-Schapira SODA`06], that achieves a competitive ratio of (n/(8n-4)).\n\nFor a single bidder, this ratio equals 1/4 and it gracefully degrades to 1/8 as the number of bidders increases.\n\nWe note that for a single bidder, online Submodular Welfare is equivalent to online Unconstrained Submodular Maximization, for which a hardness of 1/4 is known alongside an algorithm with a matching competitive ratio.\n\nTo the best of our knowledge, no competitive ratio was previously known except for the special case of a single bidder.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture: 94984580239
UID:eventx6a5a287eead9210341
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20230110T103000
DTEND;TZID=Asia/Jerusalem:20230110T113000
DTSTAMP;TZID=Asia/Jerusalem:20230110T103000
SUMMARY: colloq  talk by Omri Ben-Eliezer (MIT)  CS Lecture: Fast Algorithms for Complex Environments  at 2023-01-10 10:30:00
DESCRIPTION:Our modern life is marked by continuous interaction with huge and complex computational environments, a setting which gives rise to numerous theoretical and algorithmic challenges. Algorithms nowadays are often required to optimize objectives that may be theoretically ill-defined, on big data that is complex-structured, while maintaining computational efficiency and provable guarantees such as privacy and robustness. In this talk I will discuss some of my work developing new computational models and fast (e.g., sublinear-time or sublinear-space) algorithms for these modern settings, where data is highly structured or undergoes complex dynamics. I will focus on three representative lines of work: (i) the first systematic investigation of adversarial robustness in streaming algorithms, (ii) a new algorithmic framework for real-world social networks based on core-periphery sparsification, and (iii) active learning and testing the “geometry of data” in low-dimensional settings. Through these examples, I will demonstrate how the symbiosis between modeling and algorithm design often leads naturally to new structural insights and multidisciplinary connections.\n\nBio:\n\n Omri Ben-Eliezer is an instructor (postdoc) in applied mathematics at MIT. He received his PhD in computer science from Tel Aviv University, under the supervision of Prof. Noga Alon, and held postdoctoral positions at Weizmann Institute and Harvard University. His research focuses on algorithm design in complex environments, with specific interests including sublinear-time and streaming algorithms, large networks, robustness and privacy, and knowledge representation. For his work, Omri received several awards, including best paper awards at PODS 2020 and at CVPR 2020 Workshop on Text and Documents, the 2021 SIGMOD Research Highlight Award, and the first Blavatnik Prize for outstanding Israeli doctoral students in computer science.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 601
UID:eventx6a5a287eeada510351
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20230110T113000
DTEND;TZID=Asia/Jerusalem:20230110T123000
DTSTAMP;TZID=Asia/Jerusalem:20230110T113000
SUMMARY: pixel-club  talk by Yuval Bahat (Princeton & University of Siegon)  Pixel Club: Neural Volume Super-Resolution  at 2023-01-10 11:30:00
DESCRIPTION:Neural volumetric representations have become a widely adopted model for radiance fields in 3D scenes. These representations are fully implicit or hybrid function approximators of the instantaneous volumetric radiance in a scene, which are typically learned from multi-view captures of the scene. We investigate the new task of neural volume super-resolution - rendering high-resolution views corresponding to a scene captured at low resolution. To this end, we propose a neural super-resolution network that operates directly on the volumetric representation of the scene. This approach allows us to exploit an advantage of operating in the volumetric domain, namely the ability to guarantee consistent super-resolution across different viewing directions. To realize our method, we devise a novel 3D representation that hinges on multiple 2D feature planes. This allows us to super-resolve the 3D scene representation by applying 2D convolutional networks on the 2D feature planes. We validate the proposed method's capability of super-resolving multi-view consistent views both quantitatively and qualitatively on a diverse set of unseen 3D scenes, demonstrating a significant advantage over existing approaches. \n\nBio:\nYuval holds a joint postdoctoral researcher position at the computational imaging lab in Princeton and the ZESS center at the university of Siegen. His research interests lie at the intersection of computer vision and computational photography with Machine learning. He was previously a postdoctoral researcher at Prof. Tomer Michaeli's lab at the Technion, after completing his PhD at the Weizmann Institute of Science, advised by Prof. Michal Irani. Prior to that he completed his M.Sc. at the Technion with Prof. Yoav Y. Schechner.\n\n
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Room 337 taub bld.
UID:eventx6a5a287eeadb610355
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20230110T123000
DTEND;TZID=Asia/Jerusalem:20230110T143000
DTSTAMP;TZID=Asia/Jerusalem:20230110T123000
SUMMARY: CSpecial Event   Recruitment Day by VAYYAR  at 2023-01-10 12:30:00
DESCRIPTION:VAYYAR representatives will visit CS to present products and developments and to offer open positions, on Wednesday, January 11, 2023, 12:30, Taub lobby.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:CS Taub Lobby
UID:eventx6a5a287eeaddc10353
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20230110T143000
DTEND;TZID=Asia/Jerusalem:20230110T153000
DTSTAMP;TZID=Asia/Jerusalem:20230110T143000
SUMMARY: colloq  talk by Tamir Tassa (Open University of Israel)  CS Colloquia: Fear Not, Vote Truthfully: Secure E-Voting Protocols for Score-and Order-Based Rules  at 2023-01-10 14:30:00
DESCRIPTION:Electronic voting systems are essential for holding virtual elections, and the need for such systems increases due to the COVID-19 pandemic and the social distancing that it mandates. One of the main challenges in e-voting systems is to secure the voting process: namely, to certify that the computed results are consistent with the cast ballots, and that the privacy of the voters is preserved. We propose secure voting protocols for elections that are governed by two central families of voting rules: score-based and order-based rules. Our protocols offer perfect ballot secrecy, in the sense that they issue only the required output, while no other information on the cast ballots is revealed. Such perfect secrecy, which is achieved by employing secure multiparty computation tools, may increase the voters’ confidence and, consequently, encourage them to vote according to their true preferences. The protocols' high level of privacy, and their lightweight nature, make them an adequate and powerful tool for democracies of any size.\n \nJoint work with Lihi Dery and Avishay Yanai\n\nShort bio:\nProfessor Tamir Tassa is a faculty member in the Department of Mathematics and Computer Science at the Open University of Israel. Previously, he served as a lecturer and researcher in the School of Mathematical Sciences at Tel Aviv University, and in the Department of Computer Science at Ben Gurion University. During the years 1993-1996 he served as an assistant professor of Computational and Applied Mathematics at the University of California, Los Angeles. He earned his PhD in mathematics from Tel Aviv University in 1993. His recent research interests include secure multiparty computation, privacy-preserving data publishing and data mining, and secret sharing.\n
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Room 337 taub bld.
UID:eventx6a5a287eeadea10350
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20230110T183000
DTEND;TZID=Asia/Jerusalem:20230110T203000
DTSTAMP;TZID=Asia/Jerusalem:20230110T183000
SUMMARY: CSpecial Event   Open Source Workshop  at 2023-01-10 18:30:00
DESCRIPTION:You are invited to an Open Source workshop on open source and how contributing to open source helps professional development and gaining experience at any stage of your career, in a lecture by Michal Forg, front end developer at Gong company and manager of the largest open source community in Israel, Pull Request, on Tuesday, January 10, 2023 at 18:30 in Taub 337.\n\nPlease register in advance
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Room 337 taub bld.
UID:eventx6a5a287eeadfa10352
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20230111T123000
DTEND;TZID=Asia/Jerusalem:20230111T133000
DTSTAMP;TZID=Asia/Jerusalem:20230111T123000
SUMMARY: Theory Seminar  talk by Ilan Doron-Arad (CS, Technion)  Theory Seminar: Efficient approximation for budgeted matroid independent set, budgeted matching, and budgeted matroid intersection  at 2023-01-11 12:30:00
DESCRIPTION:Abstract: We consider the budgeted matroid independent set problem. The input is a ground set, where each element has a cost and a non-negative profit, along with a matroid over the elements and a budget. The goal is to select a subset of elements which maximizes the total profit subject to the matroid and budget constraints. Several well known special cases, where we have, e.g., a uniform matroid and a budget, or no matroid constraint (i.e., the classic knapsack problem), admit a fully polynomial-time approximation scheme (FPTAS). In contrast, already a slight generalization to the multi-budgeted matroid independent set problem has a PTAS but does not admit an efficient polynomial-time approximation scheme (EPTAS). This implies a PTAS for our problem, which is the best known result prior to our work.\n\nOur main contribution is an EPTAS for the budgeted matroid independent set problem, and a generalization of the technique for obtaining an EPTAS for budgeted matching and budgeted matroid intersection. A key idea of the scheme is to find a representative set for the instance, whose cardinality depends solely on 1/\eps,\nwhere \eps>0 is the accuracy parameter of the scheme. Our scheme enumerates over subsets of the representative set and extends each subset using Lagrangian relaxation techniques.\n\nFor a single matroid, the representative set is identified via a matroid basis minimization, which can be solved by a simple greedy approach. For matching and matroid intersection, matroid basis minimization is used as a baseline for a more sophisticated approach.\n
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 201
UID:eventx6a5a287eeae0910354
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20230111T163000
DTEND;TZID=Asia/Jerusalem:20230111T173000
DTSTAMP;TZID=Asia/Jerusalem:20230111T163000
SUMMARY: Coding_Theory_Semina  talk by Prof. Camilla Hollanti (Aalto University, Finland)  Coding Theory: Capacity of Private Information Retrieval from Coded and Colluding Servers  at 2023-01-11 16:30:00
DESCRIPTION:Private information retrieval (PIR) addresses the question of how to retrieve data items from a database or cloud without disclosing information about the identity of the data items retrieved. The area has received renewed attention in the context of PIR from coded storage. Here, the files are distributed over the servers according to a storage code instead of mere replication. Alongside with the basic principles of PIR, we will review recent capacity results and demonstrate the usefulness of the so-called star product PIR scheme.\n\nThe talk is based on joint work with Ragnar Freij-Hollanti, Oliver Gnilke,  Lukas Holzbaur, David Karpuk, and Jie Li.\n\nBio:\nCamilla Hollanti received the M.Sc. and Ph.D. degrees from the University of Turku, Finland, in 2003 and 2009, respectively, both in pure mathematics. Since 2011, she has been with the Department of Mathematics and Systems Analysis, Aalto University, Finland, where she currently works as a professor and the vice head of the department, and leads a research group in Algebra, Number Theory, and Applications. From 2017 to 2020, she was affiliated with the Institute of Advanced Studies, Technical University of Munich, where she held a Hans Fischer Fellowship. Her research interests lie within applications of algebraic number theory to wireless communications and physical layer security as well as in combinatorial and coding theoretic methods related to secure and private computation.\n\nDr. Hollanti is a coauthor of over a hundred scientific peer-reviewed publications and is a recipient of several grants, including seven Academy of Finland Grants. In 2014, she received the World Cultural Council Special Recognition Award for young researchers. In 2017, the Finnish Academy of Science and Letters awarded her the Väisälä Prize in Mathematics. Since 2020, she has been serving as a member of the Board of Governors of the IEEE Information Theory Society, and she was a general chair of the IEEE ISIT 2022. She is currently an Editor of the AIMS Journal on Advances in Mathematics of Communications, SIAM Journal on Applied Algebra and Geometry, IEEE Transactions on Information Theory, and Annales Fennici Mathematici.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture: 98686325633
UID:eventx6a5a287eeae1910330
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20230112T103000
DTEND;TZID=Asia/Jerusalem:20230112T113000
DTSTAMP;TZID=Asia/Jerusalem:20230112T103000
SUMMARY: colloq  talk by Re'em Harel (Head of Algorithms, NRCN)  CS Guest Lecture: Automatic Parallelization for Concurrent Programming – Past, Present, and Future  at 2023-01-12 10:30:00
DESCRIPTION:Introducing parallelism to applications is a complex and tedious task. As a result, the field named automatic parallelization emerged. Automatic parallelization refers to the seamless introduction of parallel schemes (such as OpenMP directives) to code. In other words, creating a tool that will mimic the human comprehension process to insert parallelization schemes. In the recent past, the main focus of this field was on creating deterministic tools such as specific functionality embedded in compilers and dedicated source-to-source (S2S) compilers. However, recent advances and success in deep Natural Language Processing (NLP) inspired models for similar code-related tasks. For example, Codex (based on GPT) generates and suggests code. The possibility of creating a similar model for automatically introducing, or at the very least suggesting, OpenMP directives rises.\n\nIn this talk, we will go through this field's history and the state-of-the-art - from the deterministic/algorithmic approach to the Machine Learning one, based on our recent publications.\n\nReferences:\n\n●	Harel, R. E., Mosseri, I., Levin, H., Alon, L. O., Rusanovsky, M., & Oren, G. (2020). Source-to-source parallelization compilers for scientific shared-memory multi-core and accelerated multiprocessing: analysis, pitfalls, enhancement and potential. International Journal of Parallel Programming, 48(1), 1-31.\n●	Mosseri, I., Alon, L. O., Harel, R. E., & Oren, G. (2020, September). ComPar: optimized multi-compiler for automatic OpenMP S2S parallelization. In International Workshop on OpenMP (pp. 247-262). Springer, Cham.\n●	Harel, R. E., Pinter, Y., & Oren, G. (2022). Learning to Parallelize in a Shared-Memory Environment with Transformers. arXiv preprint arXiv:2204.12835. Extended Abstract: The International Conference for High-Performance Computing, Networking, Storage, and Analysis (SC 2022)\n\nBiography: Re’em Harel is a computer science Ph.D. student at Ben-Gurion University and an oneAPI student ambassador. The main focus of his Ph.D. research is using state-of-the-art NLP models to automatically introduce parallelization schemes, such as OpenMP directives and MPI functions, to new and legacy codes. In addition, he is a researcher in the scientific computing lab at NRCN, focusing on parallel programming and scientific computing.\n
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture: 95409713968
UID:eventx6a5a287eeae2c10334
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20230112T123000
DTEND;TZID=Asia/Jerusalem:20230112T133000
DTSTAMP;TZID=Asia/Jerusalem:20230112T123000
SUMMARY: Theory Seminar  talk by Merav Parter (Weizmann Institute of Science)  Theory Seminar: A Graph Theoretic Approach for Resilient Distributed Algorithms  at 2023-01-12 12:30:00
DESCRIPTION:Following the immense recent advances in distributed networks, the explosive growth of the Internet, and our increased dependency on these infrastructures, guaranteeing the uninterrupted operation of communication networks has become a major objective in network algorithms. The modern instantiations of distributed networks, such as the Bitcoin network and cloud computing, introduce new security challenges that deserve urgent attention in both theory and practice.\n\nIn this talk, I will present a unified framework for obtaining fast, resilient and secure distributed algorithms for fundamental graph problems. Our approach is based on a graph-theoretic perspective in which common notions of resilient requirements are translated into suitably tailored combinatorial graph structures. We will discuss recent developments along the following two lines of research:\n\n– Initiating and establishing the theoretical exploration of security in distributed graph algorithms. Such a notion has been addressed before mainly in the context of secure multi-party computation (MPC). The heart of our approach is to develop new graph theoretical infrastructures to provide graphical secure channels between nodes in a communication network of an arbitrary topology.\n\n– Designing distributed algorithms that can handle various adversarial settings, such as, node crashes and Byzantine attacks. We will mainly provide general compilation schemes that are based on exploiting the high-connectivity of the graph. Our key focus will be on the efficiency of the resilient algorithms in terms of the number of communication rounds.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 201
UID:eventx6a5a287eeae3e10345
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20230115T133000
DTEND;TZID=Asia/Jerusalem:20230115T143000
DTSTAMP;TZID=Asia/Jerusalem:20230115T133000
SUMMARY: MSC  talk by Haitham Fadila  Kernel-based Construction Operators for Boolean Sum and Ruled Geometry  at 2023-01-15 13:30:00
DESCRIPTION:Boolean sum and ruling are two well-known construction operators for both parametric surfaces and trivariates.\n\nIn many cases, the input freeform curves in R^2 or surfaces in R^3 are complex, and as a result, these construction operators might fail to build the parametric geometry so that it has positive Jacobian throughout the domain.\n\nIn this work, we focus on cases in which those constructors fail to build parametric geometries with a positive Jacobian throughout while the freeform input has a kernel point.\n\nWe show that in the limit, for high enough degree raising or enough refinement, our construction scheme must succeed if a kernel exists.\n\nIn practice, our experiments, on quadratic, cubic and quartic Bezier and B-spline curves and surfaces show that for a reasonable degree raising and/or refinement, the vast majority of construction examples are successful.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture: 95299541427 and Taub 301
UID:eventx6a5a287eeae4e10339
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20230115T143000
DTEND;TZID=Asia/Jerusalem:20230115T153000
DTSTAMP;TZID=Asia/Jerusalem:20230115T143000
SUMMARY: Coding_Theory_Semina  talk by Dor Elimelech (Ben-Gurion university   Coding Theory: The Generalized Covering Radius of Codes  at 2023-01-15 14:30:00
DESCRIPTION:The generalized covering radius (GCR) was recently introduced as a fundamental property of linear codes, shown to characterize a trade-off between storage amount, access complexity, and latency in linear data querying protocols (such as many PIR protocols). In the general case (where the codes are not necessarily linear), the GCR is used in order to formulate a higher-order version of the famous combinatorial football-pool problem. During this talk, we shall discuss the equivalent definitions and basic properties of the GCR and survey the recent progress in the study of generalized covering codes.\n\nDor Elimelech received his B.Sc. in mathematics and his B.Sc. in electrical engineering in 2018 from Ben-Gurion University of the Negev, Israel; his M.Sc. degree in electrical engineering in 2020 from Ben-Gurion University of the Negev (summa cum laude). In 2020 he started his Ph.D. in electrical engineering, also at Ben-Gurion University, supervised by Prof. Moshe Schwartz and Prof. Tom Meyerovitch. His research interests include coding theory, probability, and dynamical systems.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 601
UID:eventx6a5a287eeae5e10356
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20230115T150000
DTEND;TZID=Asia/Jerusalem:20230115T160000
DTSTAMP;TZID=Asia/Jerusalem:20230115T150000
SUMMARY: PHD  talk by  Avi Kaplan  Computational Complexity under Communication Constraints  at 2023-01-15 15:00:00
DESCRIPTION:Can preprocessing help reduce computational costs? We study this question in the context of communication complexity, focusing on a simple "simultaneous messages" setting in which computationally unbounded Alice and Bob each send a single message to a computationally bounded Carol.\n\nA big part of our work concentrates on the task of computing the inner product function modulo 2 by a polynomial-sized bounded-depth Boolean circuit. Without preprocessing this task was shown to be impossible, and we show that this is still not possible even if we allow preprocessing limited to sublinear stretch of the inputs. Another part of our work goes beyond inner product and bounded-depth circuits, and explores other computational problems and constraints on Carol in the simultaneous messages framework.\n\nThe above question turns out to be closely related to another question, which is independently motivated by cryptographic applications. Suppose that two distributions X and Y are k-indistinguishable, in the sense that their projections to any k coordinates are identically distributed. Can some constant-depth circuit distinguish between X and Y? A celebrated theorem by Braverman implies a negative answer when X is uniform, whereas a work of Bogdanov et al. shows that this is not the case in general. We initiate a systematic study of this question for natural classes of "simple" distributions, including ones that arise in cryptographic applications, obtaining positive results, negative results, and barriers.\n
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture: 95627681547 and Taub 301
UID:eventx6a5a287eeae6d10343
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20230115T150000
DTEND;TZID=Asia/Jerusalem:20230115T160000
DTSTAMP;TZID=Asia/Jerusalem:20230115T150000
SUMMARY: PHD  talk by David Naori  New Models and Improved Bounds for Online Optimization  at 2023-01-15 15:00:00
DESCRIPTION:We extend the standard online worst-case model to accommodate past experience which is available to the online player in many practical scenarios. We do this by revealing a random sample of the adversarial input to the online player ahead of time.\n\nThe online player competes with the expected optimal value on the part of the input that arrives online. Our model bridges between existing online stochastic models (e.g., items are drawn i.i.d. from a distribution) and the online worst-case model. We also extend in a similar manner (by revealing a sample) the online random-order model.\n\nWe study the secretary problem and online matching problems in our new models. We also study the online facility location problem and obtain improved bounds in the standard random-order model.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture: 91014628593 and Taub 401
UID:eventx6a5a287eeae7e10349
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20230117T113000
DTEND;TZID=Asia/Jerusalem:20230117T123000
DTSTAMP;TZID=Asia/Jerusalem:20230117T113000
SUMMARY: pixel-club  talk by Mark Sheinin (Carnegie Mellon)  Pixel Club: Computational Imaging for Enabling Vision Beyond Human Perception  at 2023-01-17 11:30:00
DESCRIPTION:From minute surface vibrations to very fast-occurring events, the world is rich with phenomena humans cannot perceive. Likewise, most computer vision systems are primarily based on 'conventional' cameras, which were designed to mimic the imaging principle of the human eye, and therefore are equally blind to these ubiquitous phenomena. In this talk, I will show that we can capture these hidden phenomena by creatively building novel vision systems composed of common off-the-shelf components (i.e., cameras and optics) coupled with cutting-edge algorithms.\n\nSpecifically, I will cover three projects using computational imaging to sense hidden phenomena. First, I will describe the ACam - a camera designed to capture the minute flicker of electric lights ubiquitous in our modern environments. I will show that bulb flicker is a powerful visual cue that enables various applications like scene light source unmixing, reflection separation, and remote analyses of the electric grid itself. Second, I will describe Diffraction Line Imaging, a novel imaging principle that exploits diffractive optics to capture sparse 2D scenes with 1D (line) sensors. The method's applications include capturing fast motions (e.g., actors and particles within a fast-flowing liquid) and structured light 3D scanning with line illumination and line sensing. Lastly, I will present a new approach for sensing minute high-frequency surface vibrations (up to 63kHz) for multiple scene sources simultaneously, using "slow" sensors rated for only 130Hz. Applications include capturing vibration caused by audio sources (e.g., speakers, human voice, and musical instruments) and localizing vibration sources (e.g., the position of a knock on the door).\n \nBio: \nMark Sheinin is a Post-doctoral Research Associate at Carnegie Mellon University's Robotic Institute at the Illumination and Imaging Laboratory. He received his Ph.D. in Electrical Engineering from the Technion - Israel Institue of Technology in 2019. His work has received the Best Student Paper Award at CVPR 2017 and the Best Paper Honorable Mention Award at CVPR 2022. He received the Porat Award for Outstanding Graduate Students, the Jacobs-Qualcomm Fellowship in 2017, and the Jacobs Distinguished Publication Award in 2018. His research interests include computational photography and computer vision.\n
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Room 1003, EE Meyer Building
UID:eventx6a5a287eeae8e10363
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20230118T123000
DTEND;TZID=Asia/Jerusalem:20230118T133000
DTSTAMP;TZID=Asia/Jerusalem:20230118T123000
SUMMARY: Theory Seminar  talk by Tal Herman (Weizmann Institute of Science)  Theory Seminar: Verifying The unseen: Interactive proofs for Label-Invariant distribution Properties  at 2023-01-18 12:30:00
DESCRIPTION:Given i.i.d. samples from an unknown distribution over a large domain [N], approximating several basic quantities, including the distribution’s support size, its entropy, and its distance from the uniform distribution, requires (NlogN) samples [Valiant and Valiant, STOC 2011].\n\nSuppose, however, that we can interact with a powerful but untrusted prover, who knows the entire distribution (or a good approximation of it). Can we use such a prover to approximate (or rather, to approximately {\em verify}) such statistical quantities more efficiently? We show that this is indeed the case: the support size, the entropy, and the distance from the uniform distribution, can all be approximately verified via a 2-message interactive proof, where the communication complexity, the verifier’s running time, and the sample complexity are O(N) . For all these quantities, the sample complexity is tight up to \polylogN factors (for any interactive proof, regardless of its communication complexity or verification time).\n\nMore generally, we give a tolerant interactive proof system with the above sample and communication complexities for verifying a distribution’s proximity to any label-invariant property (any property that is invariant to re-labeling of the elements in the distribution’s support). The verifier’s running time in this more general protocol is also O(N) , under a mild assumption about the complexity of deciding, given a compact representation of a distribution, whether it is in the property or far from it.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 201
UID:eventx6a5a287eeae9f10358
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20230118T123000
DTEND;TZID=Asia/Jerusalem:20230118T143000
DTSTAMP;TZID=Asia/Jerusalem:20230118T123000
SUMMARY: CSpecial Event   Recruitment Day by Mobileye  at 2023-01-18 12:30:00
DESCRIPTION:Mobileye representatives will visit CS to present the development of the software and algorithms of Mobileye's autonomous vehicle, the possibilities of employment and life in the company, on Wednesday, January 18, 2023, 12:30 in the Taub lobby.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:CS Taub Lobby
UID:eventx6a5a287eeaeae10361
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20230119T103000
DTEND;TZID=Asia/Jerusalem:20230119T113000
DTSTAMP;TZID=Asia/Jerusalem:20230119T103000
SUMMARY: colloq  talk by Guy Tamir (Technology Evangelist, Intel)  CS Guest Lecture: Open Software for the Parallel, Heterogeneous Future  at 2023-01-19 10:30:00
DESCRIPTION:The race for performance and the variety of specialized workloads drives the industry to build more parallel, heterogenous, and distributed computing systems. These systems introduce multiple programming challenges. This talk will overview the driving forces, world trends, challenges, and emerging solutions. Specifically, we will overview the oneAPI Initiative and its components and benefits. We will demonstrate SYCL's new programming paradigm and more.\nBiography: Guy Tamir is a technology evangelist at Intel Software and Advanced Technology group. His main areas of interest and expertise are Artificial Intelligence, Computer vision, Video processing, and Heterogeneous, multi-accelerator parallel computing. In addition, Guy is an active YouTuber with the OpenVINO and oneAPI video channel that just passed 3 million viewers recently. Guy holds an M.Sc. (EE, Technion) and MBA (Open University). Channel link: https://youtube.com/playlist?list=PLg-UKERBljNxsCltpcXU_Haz9xQSCN_SB
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture: 97146417324
UID:eventx6a5a287eeaebb10337
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20230122T103000
DTEND;TZID=Asia/Jerusalem:20230122T113000
DTSTAMP;TZID=Asia/Jerusalem:20230122T103000
SUMMARY: colloq  talk by Michal Moshkovitz (Bosch Center & Tel-Aviv University)  CS Lecture: Building the Foundations of Explainable and Interpretable Machine Learning  at 2023-01-22 10:30:00
DESCRIPTION:Machine learning (ML) is integrated into our society, it is present in the judicial, health,  transportation, and financial systems. As the integration increases, the necessity of ML transparency increases. The fields of explainable and interpretable ML attempt to add transparency to ML: either by adding explanations to a given black-box ML model or by building a model which is interpretable and self-explanatory.\n\nDespite the importance of explainability and interpretability, their foundations are missing. Basic questions are left unanswered: How to define explainability and interpretability? Is there a tradeoff between performance and interpretability? How to evaluate the quality of explanation? In this talk we start answering these questions in the realm of supervised, unsupervised, and reinforcement learning.\n\nBio: \nMichal is a research scientist at Bosch Center for AI and a visiting researcher at Tel-Aviv University. Previously, she was a postdoctoral fellow at the Qualcomm Institute of the University of California San Diego and a postdoc at Tel-Aviv University hosted by Yishay Mansour. Her interests lie in the foundations of AI, and in the last three years she has been focused on developing the mathematical foundations of explainable machine learning.\n\nMichal received her Ph.D. from the Hebrew University and an MSc from Tel-Aviv University. During her Ph.D., Michal interned at the Machine Learning for Healthcare and Life Sciences group of IBM Research and the Foundations of Machine Learning group of Google. Michal has been selected as a 2021 EECS MIT Rising Star, the recipient of the Anita Borg scholarship from Google and the Hoffman scholarship from the Hebrew University. \n\n
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 601
UID:eventx6a5a287eeaeca10365
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20230124T123000
DTEND;TZID=Asia/Jerusalem:20230124T143000
DTSTAMP;TZID=Asia/Jerusalem:20230124T123000
SUMMARY: CSpecial Event   Projects Fair on IoT, Android, Arduino and Networks  at 2023-01-24 12:30:00
DESCRIPTION:You are invited to the CS Taub projects fair for the Winter Semester of 2023, where 30 teams of undergraduate students will present and demonstrate projects in various fields in  IoT, Android, Arduino and Networks, developed as part of the final project in the software engineering and communication networks track, most of which were carried out in collaboration with various social associations and organizations, and were intended to make a contribution to the community.\n\nThe fair will be held on Tuesday, January 24, 2023, 12:30-14:30, at the CS Taub Lobby.\n\nThe presenting posters (Heb)
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:CS Taub Lobby
UID:eventx6a5a287eeaed910362
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20230125T123000
DTEND;TZID=Asia/Jerusalem:20230125T133000
DTSTAMP;TZID=Asia/Jerusalem:20230125T123000
SUMMARY: Theory Seminar  talk by Michal Wlodarczyk (Ben-Gurion University)  Theory Seminar: Hitting Minors, Planarization, and Kernelization  at 2023-01-25 12:30:00
DESCRIPTION:The concept of a graph minor is fundamental in topological graph theory. First, I will describe the cornerstones of this theory from the lens of parameterized complexity. Next, I will survey more recent results concerning minor-hitting problems, focusing on three algorithmic paradigms: approximation, kernelization, and parameterized algorithms. Here, an important special case is the Vertex Planarization problem (remove as few vertices as possible to make a given graph planar) – this problem is equivalent to hitting all K_5 and K_{3,3} minors in a given graph. Finally, I will talk about our recent result: an O(1)-approximate kernel for Vertex Planarization, being a combination of approximation and kernelization. This is a joint work with Bart. M. P. Jansen. No prior background on graph minors or kernelization is required.\n
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 201
UID:eventx6a5a287eeaee910359
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20230126T103000
DTEND;TZID=Asia/Jerusalem:20230126T113000
DTSTAMP;TZID=Asia/Jerusalem:20230126T103000
SUMMARY: CSpecial Talk  by Tim Mattson (Senior principal engineer, Intel)  CS Guest Lecture: Software Development in the Sixth Epoch of Distributed Computing  at 2023-01-26 10:30:00
DESCRIPTION:Amin Vahdat, in a talk that has gone viral, described the five epochs of distributed computing (https://www.youtube.com/watch?v=Am_itCzkaE0).  It’s a great talk, but I disagree with him on one key point.  He thinks we are early in the fifth Epoch.  I say we entered the fifth Epoch several years ago and we are on the verge of the next Epoch … the sixth Epoch of distributed computing.\n\nIn this talk I will very briefly outline the five Epochs of distributed computing and then shift to the future and the sixth Epoch. This Epoch emerges when we bring next generation networking technology into our distributed computing systems so the time for one hop on the network is on par with the time for a memory reference in DRAM (Distributed Random Access Memory).\n\nThis innovation is coming in the not-too-distant future.  It will fundamentally change how high-performance computing applications project into the cloud.  We need to start thinking NOW about how we will develop software in the sixth Epoch.   I will suggest one approach for programming in the sixth Epoch, but the ideas are speculative and therefore alternatives abound. To that end, I hope this talk launches an aggressive, and hopefully interesting, dialog about software development in the sixth epoch of distributed computing.\n\nBiography: Tim Mattson is a parallel programmer obsessed with every variety of science (Ph.D. Chemistry, UCSC, 1985).  He is a senior principal engineer in Intel’s parallel computing lab.  Tim has been with Intel since 1993 and has worked with brilliant people on great projects including: (1) the first TFLOP computer (ASCI Red), (2) MPI, OpenMP and OpenCL, (3) two different research processors (Intel's TFLOP chip and the 48 core SCC), (4) Data management systems (Polystore systems and Array-based storage engines), and (5) the GraphBLAS API for expressing graph algorithms as sparse linear algebra. Tim has well over 150 publications including five books on different aspects of parallel computing, the latest (Published November 2019) titled “The OpenMP Common Core: making OpenMP Simple Again”.\n
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture: 94604196201
UID:eventx6a5a287eeaefa10335
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20230126T110000
DTEND;TZID=Asia/Jerusalem:20230126T120000
DTSTAMP;TZID=Asia/Jerusalem:20230126T110000
SUMMARY: MSC  talk by Ariel Larey  Develop Novel Computer Vision and Deep Learning Techniques for Digital Pathology  at 2023-01-26 11:00:00
DESCRIPTION:The diagnosis and treatment planning of many diseases, such as cancer and auto-immune conditions, rely on histological slides. In recent years, digital pathology has become more abundant allowing high-thruput digitization of pathology images and the use of AI to analyze and interpret them. Yet, there are still inherent challenges in harnessing AI for pathology that includes coping with features in multiple size scales, the ability to achieve interpretability of the AI results, and biased datasets that impede the ability to produce reliable decision systems.\n\nIn the first part of the talk, we will present our AI-based decision support system for the diagnosis of Eosinophilic esophagitis (EoE), a chronic immune disease that is second only to gastroesophageal reflux disease as the leading cause of chronic refractory dysphagia in adults and children. Diagnostics of EoE rely on counting single immune cells within a huge whole-slide image, a time-consuming process that is prone to errors. Our platform goes beyond recapturing the current manual histological gold standard by AI and reveals novel local and spatial biomarkers for EoE diagnosis, and can be harnessed to the diagnostics of other conditions.\n\nIn the second part, we will present a novel approach for generating synthetic semantic masks of histological tissues. Many histology datasets are biased due to biological factors (many healthy patients and many very sick but not enough around the decision threshold) or technical factors (images from a particular device or a specific campus). While GAN-based solutions can produce realistic textures, their ability to recapitulate the spatial distribution of features in tissues is limited. One solution is to use image translation conditional networks, however, generating proper conditional masks of tissues, as input to the network, is still not done successfully. We will present our approach that can produce realistic semantic masks of various organs such as lungs and skin. This allows the generation of synthetic histology slides by controlling their spatial distribution, thus providing unbiased datasets that can facilitate the development and testing of AI pathology solutions.\n
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture: 98200430832 and Faculty of Medicine, seminar room 4th floor
UID:eventx6a5a287eeaf0e10357
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20230130T143000
DTEND;TZID=Asia/Jerusalem:20230130T153000
DTSTAMP;TZID=Asia/Jerusalem:20230130T143000
SUMMARY: MSC  talk by Eli Gavril  TCAN: Authentication Without Cryptography on a CAN Bus Based on Nodes Location on the Bus  at 2023-01-30 14:30:00
DESCRIPTION:Vehicles possess an extraordinary amount of technological features that are meant to improve the safety and comfort of the driving experience. Those features have become so advanced that many of the driving aspects are now almost completely automated. Most drivers in the world now rely on the computer systems of the vehicle itself in order to perform even the most basic tasks, such as steering and parking.\n\nThe CAN bus is the main network used for communication between the various systems of the vehicle. As such, it is a major target for attackers who wish to break into the car. Indeed, it has been proven that attacks can be performed on the CAN bus in order to cause physical damage to the vehicle. Specifically, attackers can forge messages and send them on the CAN bus in order to impersonate certain systems of the vehicle. Securing the CAN bus has therefore become a priority in the automobile industry.\n\nIn this thesis we present TCAN, an authentication mechanism for messages on the CAN bus that does not require cryptography. TCAN ensures that the messages are sent by their alleged senders, and are not modified by other parties connected to the bus. The main idea of TCAN is to uniquely identify nodes on the bus by their physical location. To do this, we install dedicated nodes on the bus that measure reception time differences, which are correlated to the senders' location on the bus (due to the constant speed-of-light propagation on the bus).
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture: 8355062003
UID:eventx6a5a287eeaf2010360
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20230201T103000
DTEND;TZID=Asia/Jerusalem:20230201T113000
DTSTAMP;TZID=Asia/Jerusalem:20230201T103000
SUMMARY: colloq  talk by Aviad Levis (Computing and Mathematics at Caltech)  CS Lecture: Computational Imaging for Scientific Discovery: From Cloud Physics to Black Holes Dynamics  at 2023-02-01 10:30:00
DESCRIPTION:Imaging plays a key role in advancing science, from revealing the internal structure of clouds to providing the first visual evidence of a black hole. While both examples come from different imaging systems, they illustrate what can be achieved with modern computational approaches. Computational imaging combines concepts from physics, machine learning, and signal processing to reveal hidden structures at the smallest and largest of scales. In this talk, I will highlight how peeling away layers of the underlying physics leads to a spectrum of algorithms targeting new scientific discoveries. I will focus on the Event Horizon Telescope (EHT), a unique computational camera with the goal of imaging the glowing fluid surrounding supermassive black holes. In May of 2022, the EHT collaboration revealed the first images of the black hole at the center of our galaxy: Sagittarius A* (Sgr A*). These images were computationally reconstructed from measurements taken by synchronized telescopes around the globe. While images certainly offer interesting insights, looking toward the future, we are developing new computational algorithms that aim to go beyond a 2D image. For example, could we use EHT observations to recover the dynamic evolution or even the 3D structure? We tackle these challenges by integrating emerging AI concepts with physics models. Our hope is that in the not-too-distant future, these new and exciting prospects will enable scientific discovery and even provide a glimpse into the very nature of space-time itself in our galaxy's most extreme environment.Bio: Aviad Levis is a postdoctoral scholar in the Department of Computing and Mathematics at Caltech, working with Katie Bouman. Currently, as part of the Event Horizon Telescope collaboration, his work focuses on developing computational algorithms for imaging black hole dynamics. Prior to that, he received his Ph.D. (2020) from the Technion and his B.Sc. (2013) from Ben-Gurion University. Notably, his Ph.D. research into 3D remote sensing of clouds has paved the way for a novel space mission (CloudCT) funded by the ERC and led by Yoav Schechner, Ilan Koren, and Klaus Schilling. Aviad is a recipient of the Zuckerman and the Viterbi Postdoctoral Fellowships.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 601
UID:eventx6a5a287eeaf3210370
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20230201T113000
DTEND;TZID=Asia/Jerusalem:20230201T123000
DTSTAMP;TZID=Asia/Jerusalem:20230201T113000
SUMMARY: MSC  talk by Matan Yechieli  Low-Latency Blockchains with DAG Holography  at 2023-02-01 11:30:00
DESCRIPTION:Classical Proof-of-Work blockchains like Bitcoin implement a decentralized ledger, where anyone can participate. They aggregate transactions from system users in blocks and decide each block's position in the ledger. They require the block at each position to accrue votes until the probability of a decision change, due to chance or malice, is negligible. To allow consumer usage of such systems, low latency in the order of seconds is necessary. In classical blockchain systems latency is in the order of hours. Recent protocols use parallel voting and reach low latency, but require a prohibitively high overhead bandwidth. \n\nWe present Holograph, a blockchain protocol that achieves a latency of seconds under practical bandwidth limitations. To achieve this we introduce a novel technique called dag (Directed Acyclic Graph) Holography. Holograph participants form a single physical block dag. Each physical block manifests as multiple virtual blocks in multiple virtual dags that serve as parallel voting structures.By viewing the same physical dag as many virtual ones, like a holographic image viewed from different angles, we obtain more votes with the same bandwidth. \n\nWe analyze Holograph’s latency as a function of overhead bandwidth limit, compared against prior art. To the best of our knowledge, this is the first such analysis of blockchain protocols. Our simulation shows that Holograph reaches the lowest latency across the tested bandwidth range, improving latency by about 5x compared to the state-of-the-art when bandwidth is limited. We run a prototype implementation in an emulated network reaching a latency of 7 seconds with 10kbps overhead bandwidth, which is 45% lower than the state-of-the-art with an order-of-magnitude lower, practical overhead throughput. 
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture: 93583582399 and Taub 401
UID:eventx6a5a287eeaf4410368
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20230202T150000
DTEND;TZID=Asia/Jerusalem:20230202T160000
DTSTAMP;TZID=Asia/Jerusalem:20230202T150000
SUMMARY: CSpecial Talk  by Hagit Attiya (CS, Technion) and Constantin Enea (École Polytechnique/CNRS)  EuroTech: Using Concurrent Objects in Randomized Programs  at 2023-02-02 15:00:00
DESCRIPTION:Atomic concurrent objects, whose operations take place instantaneously, are a powerful technique for designing complex concurrent programs. Since they are not always available, they are typically substituted with software implementations. A prominent condition relating these implementations to their atomic specifications is linearizability, which preserves safety properties of programs using them. However linearizability does not preserve hyper-properties, which include probabilistic guarantees about randomized programs. A more restrictive property, strong linearizability, does preserve hyper-properties but it is impossible to achieve in many situations.\n\nIn particular, we show that there are no strongly linearizable implementations of multi-writer registers or snapshot objects in message-passing systems. On the other hand, we show that a wide class of linearizable implementations, including well-known ones for registers and snapshots, can be modified to approximate the probabilistic guarantees of randomized programs when using atomic objects.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture:\n\nRegistration
UID:eventx6a5a287eeaf5510367
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20230207T103000
DTEND;TZID=Asia/Jerusalem:20230207T113000
DTSTAMP;TZID=Asia/Jerusalem:20230207T103000
SUMMARY: colloq  talk by Nadav Amit (VMware Research)  CS Lecture: Reexamining Basic OS Memory Management Techniques  at 2023-02-07 10:30:00
DESCRIPTION:Despite significant advancements in operating system memory management, our understanding of the desired behavior of fundamental techniques introduced decades ago is sometimes incomplete or not well-defined. This can result in correctness issues that might cause the system to crash or be compromised, as well as missed opportunities for optimizations. In this talk, I will present two specific examples of this: (1) the inefficiencies in synchronizing the memory view across different CPU cores, and (2) the undefined behavior of the interactions between two common memory management mechanisms, copy-on-write and pinning.\nBio:Nadav Amit is a senior researcher in VMware Research. He received his PhD in 2014 from the Technion, Israel Institute of Technology for his work on alleviating of virtualization bottlenecks. He is a recipient of the SPEC Distinguished Dissertation Award, IBM Fellowship Award and an honorable mention of the Dennis M. Ritchie Doctoral Dissertation Award. His current research revolves operating systems and virtualization and focuses on memory management.\n\n
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 601
UID:eventx6a5a287eeaf6510373
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20230208T153000
DTEND;TZID=Asia/Jerusalem:20230208T163000
DTSTAMP;TZID=Asia/Jerusalem:20230208T153000
SUMMARY: MSC  talk by Almog Zur  RELAX: Recovering Lazily from Failed Execution with
Persistent Memory  at 2023-02-08 15:30:00
DESCRIPTION:Recent non-volatile main memory technology (such as Intel’s Optane) gave rise to an abundance of research on building persistent data structures, whose content can be recovered after a system crash. While there has been significant progress in making durable data structures efficient, shortening the length of the recovery phase after a crash (in which data cannot be accessed) has not received much attention. In fact, programmers need to choose exclusively between durable data structures that provide high performance during normal (failure-free) execution and durable data structures that provide fast recovery from crashes. In this paper we present the RELAX general transformation. RELAX generates durable data structures that provide the best of both worlds. They provide high performance with almost zero recovery time, with an overhead that quickly descends following a crash event, until the program soon regains maximal performance. We implemented RELAX on a hash table, a skip list, a binary tree, a linked list, and an array. The evaluation shows that the generated data structures are fast, and that following a crash, even large data sets with hundreds of millions of nodes become responsive within a few milliseconds, whereas other efficient constructions require more than 10 seconds of unresponsive recovery time.\n
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 601
UID:eventx6a5a287eeaf7510364
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20230212T113000
DTEND;TZID=Asia/Jerusalem:20230212T123000
DTSTAMP;TZID=Asia/Jerusalem:20230212T113000
SUMMARY: MSC  talk by Gilad Chase  Generalized polymorphisms  at 2023-02-12 11:30:00
DESCRIPTION:We determine all $m$-ary Boolean functions $f_0,\ldots,f_m$ and $n$-ary Boolean functions $g_0,\ldots,g_n$ satisfying the equation $f_0(g_1(z_{11},\ldots,z_{1m}),\ldots,g_n(z_{n1},\ldots,z_{nm})) = g_0(f_1(z_{11},\ldots,z_{n1}),\ldots,f_m(z_{1m},\ldots,z_{nm})),$ for all Boolean inputs $\{ z_{ij} : i \in [n], j \in [m] \}$. This extends characterizations by Dokow and Holzman[DH09] (who considered the case $g_0 = \cdots = g_n$) and by Chase, Filmus, Minzer, Mossel and Saurabh [CFMMS22] (who considered the case $g_1 = \cdots = g_n$).
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture: 92374147324 and Taub 301
UID:eventx6a5a287eeaf8510371
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20230212T123000
DTEND;TZID=Asia/Jerusalem:20230212T133000
DTSTAMP;TZID=Asia/Jerusalem:20230212T123000
SUMMARY: PHD  talk by Aviv A. Rosenberg  One sequence, one structure? Computationally identifying protein structures that defy the central dogma of biology  at 2023-02-12 12:30:00
DESCRIPTION:Proteins fold from a sequence of amino acids, forming secondary structures which subsequently fold into a three-dimensional structure that enables their function. The amino acid sequence is defined in the genetic sequence as codons,  many of which are synonymous, i.e., they code for the same amino acid. The "one sequence, one structure" dogma, established over half a century ago, remains the commonly accepted notion, and implies that synonymous coding is inconsequential to protein structure. This talk will present results from three different works, which challenge this dogma through large-scale computational analysis of protein structures.First, we develop novel methods for computing and comparing codon-specific protein backbone angle distributions. We design a non-parametric approach for comparing these bivariate distributions using finite samples, and identify synonymous codon distributions which are distinguishable, with statistical significance, within some secondary structures. This demonstrates, for the first time, an association between synonymous codon usage and the final protein structure around the amino acids they translate into.Next, we expand this approach to consider pairs of amino acids, accounting for the peptide bond which is formed between amino acids during translation of the genetic code. To that end, we introduce a tool for defining local, two amino acid-long sub-secondary structural units. We analyze the joint distribution of backbone angles across the peptide bond and show that our structural units can more meaningfully represent backbone conformations than conventional secondary structure.Finally, building on the aforementioned tools, we devise a constructive approach for pinpointing locations in highly similar protein structures having vastly different local backbone conformations despite residing in environments with an identical sequence and potential interaction network. We show that such conformational differences are stable under molecular dynamics simulations, and that they are not predicted by AlphaFold, a state-of-the-art structure prediction model which relies only on the amino acid sequence. Our data-driven approach provides biologists with invaluable dogma-defying examples, guiding further research into the mechanisms behind protein folding.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture: 97521197354 and Taub 601
UID:eventx6a5a287eeaf9410366
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20230215T123000
DTEND;TZID=Asia/Jerusalem:20230215T133000
DTSTAMP;TZID=Asia/Jerusalem:20230215T123000
SUMMARY: Theory Seminar  talk by Seth pettie (University of Michigan)  Theory seminar: Algorithms Should Have Bullshit Detectors! (or Polynomial Time Byzantine Agreement with Optimal Resilience)  at 2023-02-15 12:30:00
DESCRIPTION:One thing that distinguishes (theoretical) computer science from other scientific disciplines is its full-throated support of a fundamentally adversarial view of the universe. Malicious adversaries, with unbounded computational advantages, attempt to foil our algorithms at every turn and destroy their quantitative guarantees. However, there is one strange exception to this world view and it is this: the algorithm must accept its input as sacrosanct, and may never simply reject its input as illegitimate. But what if some inputs really are illegitimate? Is building a “bullshit detector” for algorithms a good idea?\n\nTo illustrate the power of the Bullshit Detection worldview, we give the first polynomial-time protocol for Byzantine Agreement that is resilient to f < n/3 corruptions against an omniscient, computationally unbounded adversary in an asynchronous message-passing model. (This is the first improvement to Ben-Or and Bracha’s exponential time protocols from the 1980s that are resilient to f < n/3 corruptions.) The key problem is to design a coin-flipping protocol in which corrupted parties (who chose the outcomes of their coins maliciously) are eventually detected via statistical tests.\n\nWe will also discuss other algorithmic contexts in which bullshit detection might be useful.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 201
UID:eventx6a5a287eeafa710374
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20230216T113000
DTEND;TZID=Asia/Jerusalem:20230216T123000
DTSTAMP;TZID=Asia/Jerusalem:20230216T113000
SUMMARY: cggc  talk by Rephael wenger (The ohio State University)  CGGC Seminar: Intro to Morse Theory, Morse-Smale Complexes, and Discrete Morse Theory  at 2023-02-16 11:30:00
DESCRIPTION:Rephael Wenger is a professor in the computer science and engineering department of The Ohio State university where he works on geometric modeling, mesh generation, geometric algorithms and scientific visualization.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 401
UID:eventx6a5a287eeafb710375
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20230315T113000
DTEND;TZID=Asia/Jerusalem:20230315T123000
DTSTAMP;TZID=Asia/Jerusalem:20230315T113000
SUMMARY: MSC  talk by Yaron Hay  A Coloring-based Approach for Concurrent Execution of Transactions and Smart Contracts in Active Replication and Blockchain Systems  at 2023-03-15 11:30:00
DESCRIPTION:Blockchain Networks, especially those with Smart Contracts, are well-known examples of Active Replication Systems. Active Replication Services are available thanks to a group of servers called replicas that handle client requests. Each server maintains a local copy of the global state of the service, and all servers update their local copy at synchronized incremental steps. At the i-th step, all servers receive the *same* transaction from a global ordering service, execute it and apply the results to their local copies. The (i+1)-th step repeats this process for the next transaction.\n\nThe combination of a consistent global ordering protocol and having all transactions are inherently deterministic guarantees that all replicas will eventually have the same exact state (for each step taken). It is imperative for replicated services. However, this comes at a loss in performance because of the restriction to executing only one transaction at any given time. Sequential execution prevents us from utilizing the capabilities of multicore processors by processing multiple transactions in parallel.\n\nIn this lecture, we will define formal models that allow for consistent concurrent execution while maintaining deterministic results at all replicas, and discuss ways to maximize concurrency.\n\nWe'll explore the connection between Graph Vertex Coloring and minimizing overall latency. Then describe practical approaches to solving this problem in real-world applications, with experimental evaluation for two popular benchmarking frameworks.\n\nLastly, we'll describe how to implement concurrent execution while replacing lock-based synchronization primitives with cheap signals for communication.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 301
UID:eventx6a5a287eeafc410376
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20230315T113000
DTEND;TZID=Asia/Jerusalem:20230315T123000
DTSTAMP;TZID=Asia/Jerusalem:20230315T113000
SUMMARY: PHD  talk by Shir Cohen  Distributed Services Under Attack  at 2023-03-15 11:30:00
DESCRIPTION:For my PhD thesis seminar, I will be presenting two of my works related to the security and reliability of distributed services in the face of Byzantine attacks. In the first work “Not a COINcidence: Sub-Quadratic Asynchronous Byzantine Agreement WHP” (DISC’20), I present a solution for binary Byzantine Agreement (BA) in asynchronous systems, using a shared coin algorithm based on a VRF and VRF-based committee sampling. My algorithms work against a delayed-adaptive adversary with a word complexity of Õ(n) and O(1) expected time, breaking the O(n²) bit barrier for asynchronous Byzantine Agreement. This work is then used to solve the multivalued version of the BA problem.\n\nIn the second work “Tame the Wild with Byzantine Linearizability: Reliable Broadcast, Snapshots, and Asset Transfer” (DISC’21), I introduce the concept of Byzantine linearizability and study Byzantine-tolerant emulations of various objects from registers, including reliable broadcast, atomic snapshot, and asset transfer. This work proves that there is an f-resilient implementation of such objects from registers with n processes for f
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 601
UID:eventx6a5a287eeafd510377
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20230316T103000
DTEND;TZID=Asia/Jerusalem:20230316T113000
DTSTAMP;TZID=Asia/Jerusalem:20230316T103000
SUMMARY: MSC  talk by Noa Schiller  Efficient and Resilient Algorithms for Asynchronous Mixed Models and Decentralized Optimization  at 2023-03-16 10:30:00
DESCRIPTION:We study a hybrid distributed model, which combines message-passing and shared-memory communication layers, and investigate the minimal number of failures that can partition such systems. We prove that this number precisely captures the resilience that can be achieved by algorithms that implement a variety of shared objects and solve common tasks, like approximate agreement. \n\nIn the cluster-based model, processes are partitioned into disjoint clusters. We solve the approximate agreement problem in this model, and its generalization to higher dimensions, multidimensional approximate agreement.\n\nThen, we turn our attention to Stochastic Gradient Decent (SGD) algorithms. SGD is widely used for approximating the minimum of a cost function Q, a core part of optimization and learning algorithms.\n\nFor a strongly convex Q, our algorithm can withstand partitions of the system, and provides convergence rate that is the maximal distributed acceleration over the optimal convergence rate of sequential SGD.\n\nFor arbitrary smooth functions, the algorithm obtains the same convergence rate as sequential SGD, up to a logarithmic factor. This is achieved by using, at each iteration, a multidimensional approximate agreement algorithm. We complement this result with a lower bound showing that under system partition, some non-convex functions cannot be optimized using a distributed algorithm.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 301
UID:eventx6a5a287eeafe410379
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20230316T110000
DTEND;TZID=Asia/Jerusalem:20230316T120000
DTSTAMP;TZID=Asia/Jerusalem:20230316T110000
SUMMARY: MSC  talk by Era Choshen  ContraBAR: Contrastive Bayes-Adaptive Deep RL  at 2023-03-16 11:00:00
DESCRIPTION:In meta reinforcement learning (meta RL), an agent seeks a Bayes-optimal policy – the optimal policy when facing an unknown task that is sampled from some known task distribution. Previous approaches tackled this problem by inferring a belief over task parameters, using variational inference methods. Motivated by recent successes of contrastive learning approaches in RL, such as contrastive predictive coding (CPC), we investigate whether contrastive methods can be used for learning Bayes-optimal behavior. We begin by proving that representations learned by CPC are indeed sufficient for Bayes optimality. Based on this observation, we propose a simple meta RL algorithm that uses CPC in lieu of variational belief inference. Our method, ContraBAR, achieves comparable performance to state-of-the-art in domains with state-based observation and circumvents the computational toll of future observation reconstruction, enabling learning in domains with image-based observations. It can also be combined with image augmentations for domain randomization and used seamlessly in both online and offline meta RL settings.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 601
UID:eventx6a5a287eeaff510378
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20230322T113000
DTEND;TZID=Asia/Jerusalem:20230322T123000
DTSTAMP;TZID=Asia/Jerusalem:20230322T113000
SUMMARY: MSC  talk by Arad Kotzer  Braess Paradox in Blockchain Layer-2 Payment Networks  at 2023-03-22 11:30:00
DESCRIPTION:Layer-2 is a popular approach to deal with the scalability limitation of blockchain networks. It allows users to execute transactions without committing them to the blockchain by relying on predefined payment channels. Users together with the payment channels form a graph known as the offchain network topology. Transactions between pairs of users without a connecting channel are also supported through a path of multiple channels. Serving such transactions involves fees paid to intermediate users. In this work, we uncover the potential existence of the Braess paradox in payment networks: Sometimes, establishing a new payment channel can increase the fees paid for serving some fixed transactions. We study conditions for the paradox to appear and provide indications for the appearance of the paradox based on real data of Bitcoin's Lightning, a popular layer-2 network. Last, we discuss methods to mitigate the paradox upon establishing a new payment channel.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture: 98170371363 and Taub 401
UID:eventx6a5a287eeb00610382
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20230322T123000
DTEND;TZID=Asia/Jerusalem:20230322T133000
DTSTAMP;TZID=Asia/Jerusalem:20230322T123000
SUMMARY: Theory Seminar  talk by Idan Shabat (Ben-Gurion University)  Theory Seminar: Linear-Size Distance Oracles and Interactive Graph Structures  at 2023-03-22 12:30:00
DESCRIPTION:Given an undirected weighted graph, a distance oracle is a data structure that answers distance queries in the graph within a short time. A path-reporting distance oracle (PRDO) is a distance oracle that is also required to return a shortest path between the queried vertices. A particular interest is in oracles that have a linear storage size in the number of vertices of the input graph, and also have small query time and a good approximation factor (called the stretch).\n\nThroughout the years, there was a major progress in the results for non-path-reporting distance oracles, and an almost optimal trade-off between size, query time and stretch was achieved (“optimal” up to the widely believed girth conjecture by Erdos). However, for PRDOs, such near-optimal results were never accomplished.\n\nIn our work, we construct PRDOs with almost the same near-optimal trade-off as of non-path-reporting distance oracles. On our way to these results, we introduce notions of interactive graph structures, that capture the functionality of both PRDOs and some other combinatorial structures, such as spanners, emulators and distance preservers.\n\nA joint work of Michael Elkin and Idan Shabat.\n
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 201
UID:eventx6a5a287eeb01710386
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20230323T133000
DTEND;TZID=Asia/Jerusalem:20230323T143000
DTSTAMP;TZID=Asia/Jerusalem:20230323T133000
SUMMARY: pixel-club  talk by Jianbo Shi (University of Pennsylvania)  Pixel Club: Mental Model in Escape Room: a First-Person POV  at 2023-03-23 13:30:00
DESCRIPTION:In an Escape-room setting, we study different human behaviors when completing time-constrained tasks involving sequential decision-making and actions. We aim to construct a human mental model linking attention, episodic memory, and hand-object interaction.\n\nWe record from two egocentric cameras: a head-mounted camera and Gaze-tracking glasses. We also record from up to four third-person cameras. Additionally, we created a detailed 3D map of the room. In this talk, I will discuss the progress we have made and the challenges we faced.\n\nRegarding computer vision research, we ask: 1) Can visual synthesis benefit Object Recognition/Affordance prediction?  2) Can we build a superhuman model from observations and simulation?\n\nBio:\n\nJianbo studied Computer Science and Mathematics as an undergraduate at Cornell University where he received his B.A. in 1994. He received his Ph.D. degree in Computer Science from University of California at Berkeley in 1998, for his thesis on Normalize Cuts image segmentation algorithm. He joined The Robotics Institute at Carnegie Mellon University in 1999 as a research faculty. Since 2003, he has been with the Department of Computer & Information Science at the University of Pennsylvania.\n\nJianbo's group is developing vision algorithms for both human and image recognition. Their ultimate goal is to develop computation algorithms to understand human behavior and interaction with objects in video, and to do so at multiple levels of abstractions: from the basic body limb tracking, to human identification, gesture recognition, and activity inference. Jianbo and his group are working to develop a visual thinking model that allows computers to understand their surroundings and achieve higher-level cognitive abilities such as machine memory and learning.\n
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Room 1063, EE Meyer Building
UID:eventx6a5a287eeb02810387
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20230327T100000
DTEND;TZID=Asia/Jerusalem:20230327T120000
DTSTAMP;TZID=Asia/Jerusalem:20230327T100000
SUMMARY: CSpecial Event  Amazon Research, Alexa Shopping 2023 Internship Program Introduction  at 2023-03-27 10:00:00
DESCRIPTION:Amazon Research, Alexa Shopping 2023 Internship Program Introduction session will take place on Monday. March 27, 10:00-11:30 on Floor 2, CS Taub Building, and will present its research challenges and 2023 research internship program for graduate students in CS/EE/IEM: \n\nProgram:\n10:00-10:20 -&nbsp;Introduction to the 2023 internship program and Alexa Shopping domain overview, Liane Lewin-Eytan, Sr Mgr., Alexa Shopping, Amazon\nPresentation: “Alexa – Using AI to understand, satisfy, and predict users’ shopping needs \n10:20-10:35 -&nbsp;Alex Libov, Science Mgr., Alexa Shopping, Amazon \n10:35-10:50 -&nbsp;Avihai Mejer, Science Mgr., Alexa Shopping, Amazon \n10:50-11:30 -&nbsp;Q&amp;As session with Science &amp; Recruiting teams\n\nNo prior knowledge required and participation in the event is restricted to enrolled graduate students in CS/EE/IEM and faculty members.\n\nPlease pre-register.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Floor 2, CS Taub Building
UID:eventx6a5a287eeb03a10381
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20230328T113000
DTEND;TZID=Asia/Jerusalem:20230328T123000
DTSTAMP;TZID=Asia/Jerusalem:20230328T113000
SUMMARY: pixel-club  talk by Michael Zibulevsky (CS, Technion)  Pixel Club: Radiation Design in Computed Tomography via Convex Optimization  at 2023-03-28 11:30:00
DESCRIPTION:Proper X-ray radiation design (via dynamic fluence field modulation, FFM) allows reducing effective radiation dose in computed tomography without compromising image quality. It takes into account patient anatomy, radiation sensitivity of different organs and tissues, and location of regions of interest. We account for all these factors within a general convex optimization framework. \n\nJoint work with Anatoli Juditsky and  Arkadi Nemirovski\n\nShort bio:\nMichael Zibulevsky received his BS-MS in Electrical Engineering from MIIT - Moscow Institute of Transportation Engineering,  and PhD in Operations Research from the Technion - Israel Institute of Technology. He is currently with the Dept. of Computer Science at the Technion. Michael Zibulevsky is one of the founders of Sparse Component Analysis. His research interests include numerical methods of optimization, sparse signal representations, deep neural networks and their applications in signal/image processing and inverse problems.  \n
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture: 184212013
UID:eventx6a5a287eeb04a10384
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20230329T113000
DTEND;TZID=Asia/Jerusalem:20230329T123000
DTSTAMP;TZID=Asia/Jerusalem:20230329T113000
SUMMARY: MSC  talk by Theo J. Adrai  Deep Optimal Transport: A Practical Algorithm for Photorealistic Image Restoration  at 2023-03-29 11:30:00
DESCRIPTION:In image restoration, traditional supervised methods that seek to restore the source enjoy exceptional distortion performance but lack visual quality. With the emergence of powerful generative algorithms, many approaches focus on image realism and diversity, but forsake faithfulness to the source. Motivated by recent theoretical findings, we present a practical algorithm that optimizes source fidelity while aiming for photo-realistic results. Our method optimally transports the distribution of MMSE estimate to the natural image distribution using a simple patch-level deep representation (as simple as an auto-encoder). The results of our experiments demonstrate that we can effectively improve the perceptual quality of MMSE estimates on severe degradations.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture: theadthechnion
UID:eventx6a5a287eeb05a10385
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20230329T163000
DTEND;TZID=Asia/Jerusalem:20230329T173000
DTSTAMP;TZID=Asia/Jerusalem:20230329T163000
SUMMARY: Coding_Theory_Semina  talk by Klim Efremenko (Ben-Gurion University)  Coding Theory: Binary Codes with Resilience Beyond 1/4 via Interaction  at 2023-03-29 16:30:00
DESCRIPTION:In the reliable transmission problem, a sender, Alice, wishes to transmit a bit-string x to a remote receiver, Bob, over a binary channel with adversarial noise. The solution to this problem is to encode x using an error-correcting code. As it is long known that the distance of binary codes is at most 1/2, reliable transmission is possible only if the channel corrupts (flips) at most a 1/4-fraction of the communicated bits.\n\nWe revisit the reliable transmission problem in the two-way setting, where both Alice and Bob can send bits to each other. Our main result is the construction of two-way error-correcting codes that are resilient to a constant fraction of corruptions strictly larger than 1/4. Moreover, our code has a constant rate and requires Bob to only send one short message.\n\nCuriously, our new two-way code requires a fresh perspective on classical error-correcting codes: While classical codes have only one distance guarantee for all pairs of codewords (i.e., the minimum distance), we construct codes where the distance between a pair of codewords depends on the ``compatibility'' of the messages they encode. We also prove that such codes are necessary for our result.\n\nJoint work with Gillat Kol, Raghuvansh R. Saxena, and Zhijun Zhang.\n
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 601
UID:eventx6a5a287eeb06b10389
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20230330T103000
DTEND;TZID=Asia/Jerusalem:20230331T213000
DTSTAMP;TZID=Asia/Jerusalem:20230330T103000
SUMMARY: CSpecial Event  CS-Hackathon 2023 - Doing Good  at 2023-03-30 10:30:00
DESCRIPTION:You are invited to join the CS Hackathon-Doing Good programming competition, in collaboration with Ruth Rappaport Children&#39;s Hospital, Rambam Health Care Campus, to be held on Thursday-Friday, March30-31, 2023, at CS Taub Building, and which this year will find solutions that enhance the well-being of children and their families during hospital treatment, improve the experience of hospital staff involved in the care of these children and their families.\n\nParticipants will have industry-leading mentors, entrepreneurs and researchers who will accompany them through the process, from the concept stage, through the planning and development stage to the presentation stage, which allows them to take part in the competition without prior familiarity with the field. In addition, it will be possible to develop an independent idea or work on a challenge from the list of challenges, to and  workshops lists, and according to the rules of the game.\n\nIn preparation for the CS Doing Good faculty hackathon, there will be mandatory meetings:\nMarch 22 (Wednesday afternoon) - Rambam Hospital tour including transportation (details coming soon)\nMarch 26, 17:30 - preparatory meeting, presentation of the game rules and questions and answers in the PianoAuditoriums of Taub buildingThe prize-winning competition will last about 30 hours, registration is open and the number of places is limited.\n\n\n
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:CS Taub Building
UID:eventx6a5a287eeb07d10380
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20230402T183000
DTEND;TZID=Asia/Jerusalem:20230402T203000
DTSTAMP;TZID=Asia/Jerusalem:20230402T183000
SUMMARY: CSpecial Event  Practical Workshop by CYE  at 2023-04-02 18:30:00
DESCRIPTION:You are invited to a practical testing workshop by CYE, supervised byTal Sihonov, director of the development group at the CYE, who will talk about the importance of tests in the industry and the world of development, with an emphasis on SaaS systems and will practice techniques of writing effective tests in Python, on Sunday, April 2, 2023 at 18:30, in Taub 337. \n\nParticipating in the workshop requires prerequisite of completing the Introduction to Ssystems Programming course or practical experience from a previous job of writing code in Python.\n\nPlease pre-register - the number of places is limited!\n
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Room 337 taub bld.
UID:eventx6a5a287eeb09010388
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20230403T100000
DTEND;TZID=Asia/Jerusalem:20230403T110000
DTSTAMP;TZID=Asia/Jerusalem:20230403T100000
SUMMARY: PHD  talk by Gal Yehuda  Connections between Machine Learning and Theoretical Computer Science  at 2023-04-03 10:00:00
DESCRIPTION:We present connections between machine learning and theoretical computer science. In particular: hardness of data-set generation for deep learning problems and connections between randomness and computation in deep learning. \n
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 601
UID:eventx6a5a287eeb09f10390
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20230403T113000
DTEND;TZID=Asia/Jerusalem:20230403T123000
DTSTAMP;TZID=Asia/Jerusalem:20230403T113000
SUMMARY: pixel-club  talk by Rebecca Willet (University of Chicago)  Pixel Club: Machine Learning and Data Assimilation in the Natural Sciences and Engineering  at 2023-04-03 11:30:00
DESCRIPTION:The potential for machine learning to revolutionize scientific and engineering research is immense, but its transformative power cannot be fully harnessed through the use of off-the-shelf tools alone. To unlock this potential, novel methods are needed to integrate physical models and constraints into learning systems, accelerate simulations, and quantify model prediction uncertainty. In this presentation, we will explore the opportunities and emerging tools available to address these challenges in the context of inverse problems, data assimilation, and reduced order modeling. By leveraging ideas from statistics, optimization, scientific computing, and signal processing, we can develop new and more effective machine learning methods that improve predictive accuracy and computational efficiency in the natural sciences.\n \nShort Bio:\nProfessor of Statistics and Computer Science & Director of AI at the Data Science Institute, with a courtesy appointment at the Toyota Technological Institute at Chicago. Faculty lead of AI+Science Postdoctoral Fellow program.\n \nProf. Willett completed her Ph.D. in Electrical and Computer Engineering at Rice University in 2005 and was an Assistant then tenured Associate Professor of Electrical and Computer Engineering at Duke University from 2005 to 2013. She was an Associate Professor of Electrical and Computer Engineering, Harvey D. Spangler Faculty Scholar, and Fellow of the Wisconsin Institutes for Discovery at the University of Wisconsin-Madison from 2013 to 2018. Willett has also held visiting researcher or faculty positions at the University of Nice in 2015, the Institute for Pure and Applied Mathematics at UCLA in 2004, the University of Wisconsin-Madison 2003-2005, the French National Institute for Research in Computer Science and Control (INRIA) in 2003, and the Applied Science Research and Development Laboratory at GE Healthcare in 2002.\n
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub  012 (Learning Center Auditorium)
UID:eventx6a5a287eeb0ad10394
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20230403T173000
DTEND;TZID=Asia/Jerusalem:20230403T193000
DTSTAMP;TZID=Asia/Jerusalem:20230403T173000
SUMMARY: CSpecial Event  Meeting on: "Democracy, Hi-tech and the Future Generation"  at 2023-04-03 17:30:00
DESCRIPTION:You are invited to a meeting with hi-tech executives on "Democracy, Hi-tech and the Future Generation", with the participation of:\nDr. Kira Radinsky\nNadir Izrael\nDr. Orna Berry\nMolly Aden\nYoram Yaacovi\nOri Hadomi\nDr. YonathanYaniv\nMaor Farid\nAvner Rothschild\nOn Monday, April 3, 2023, 17:30, in Taub 1.\n\nPlease register in advance
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:CS Taub Build. Auditorium 1
UID:eventx6a5a287eeb0be10395
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20230404T143000
DTEND;TZID=Asia/Jerusalem:20230404T153000
DTSTAMP;TZID=Asia/Jerusalem:20230404T143000
SUMMARY: MSC  talk by Ella Sheory  Exploring Advanced Cache Algorithms for the TLB  at 2023-04-04 14:30:00
DESCRIPTION:The translation lookaside buffer (``TLB’’) is a small cache that accelerates virtual to physical address translation, which processors typically manage with variants of the least recently used (``LRU’’) algorithm. Although LRU is simple, it is suboptimal for some workloads. Our analysis shows that if the processor uses the optimal---but impractical---Belady algorithm instead of LRU, runtime improves by up to 15% (and 5% on average) over LRU in single-thread (``ST’’) mode. Runtime further improves by up to 23% (yet still 5% on average) in simultaneous multithreading (``SMT’’) mode, where the TLB is competitively shared between two threads.\n\nGiven this background, we observe that while past research developed various practical caching algorithms to outperform LRU, such research was typically evaluated in the context of regular data caches rather than the TLB. Our goal in this study is therefore to investigate whether advanced practical caching algorithms improve TLB performance and, if so, to what extent they can approach Belady performance.\n\nTo this end, we classify existing caching algorithms into three groups based on their additional storage requirements: small (less than 10% of the LRU-managed TLB), medium (between 10% and 50%), and high (more than 50%). From each group, we select a representative algorithm. We find that the representatives of the small and medium groups improve performance by only 1%, on average (and no more than 11%). We also find that the third algorithm's complexity and sophistication are unjustified, as we can achieve the aforementioned optimal Belady improvement by handing the associated additional storage to the simpler LRU baseline. We thus conclude that no existing practical caching algorithm can meaningfully improve LRU TLB performance.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 301
UID:eventx6a5a287eeb0cd10391
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20230404T150000
DTEND;TZID=Asia/Jerusalem:20230404T160000
DTSTAMP;TZID=Asia/Jerusalem:20230404T150000
SUMMARY: Coding_Theory_Semina  talk by Prof. Antonia Wachter-Zeh (University of Munich)  Coding Theory: Interleaved (Rank-Metric) Codes for Cryptography  at 2023-04-04 15:00:00
DESCRIPTION:Public-key cryptography is the foundation for establishing secure communication between multiple parties. Traditional public-key algorithms such as RSA are based on the hardness of factoring large numbers or the discrete logarithm problem, but can be attacked in polynomial time once a capable quantum computer exists. Code-based public-key cryptosystems are considered to be post-quantum secure, but compared to RSA or elliptic curve cryptography their crucial drawback is the significantly larger key size. In order to reduce key sizes, (interleaved) rank-metric codes can be used in code-based cryptography.\n\nIn this talk, we first give an overview of interleaving and decoding algorithms in the Hamming and rank metric and then present different approaches to define code-based cryptographic schemes. \n \nAntonia Wachter-Zeh is an Associate Professor at the Technical University of Munich (TUM), Munich, Germany in the School of Computation, Information and Technology. She received the M.Sc. degree in communications technology in 2009 from Ulm University, Germany. She obtained her Ph.D. degree in 2013 from Ulm University and from Universite de Rennes 1, Rennes, France. From 2013 to 2016, she was a postdoctoral researcher at the Technion—Israel Institute of Technology, Haifa, Israel, and from 2016 to 2020 a Tenure Track Assistant Professor at TUM. She is a recipient of the DFG Heinz Maier-Leibnitz-Preis and of an ERC Starting Grant. She is currently an Associate Editor for the IEEE Transactions on Information Theory. Her research interests are coding theory, cryptography and information theory and their application to storage, communications, privacy, security and machine learning. \n
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 601
UID:eventx6a5a287eeb0de10393
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20230419T123000
DTEND;TZID=Asia/Jerusalem:20230419T143000
DTSTAMP;TZID=Asia/Jerusalem:20230419T123000
SUMMARY: CSpecial Event  CS Open Day for Graduate Studies  at 2023-04-19 12:30:00
DESCRIPTION:Technion CS open day 2023 invites outstanding undergraduates from all universities to learn about the Computer Science Department and register for Winter Semester 2023-24.\n\nThe event will be held on Wednesday, April 19, 2022. between 12:30-14:00, room 337, Taub Building for Computer Science, Technion.\n\nThe program will include review on curriculum, research and life at the Technion CS Department:\n- CS Dean, Prof. Danny Raz\n- Vice Dean, Prof. Gill Barequet\n- Dr. Liane Levy-Eitan, Director Research Group at Amazon\n- Mr. Dean Zadok (Ph.D. students) \n- Questions and answers\n\nFor more information please contact Graduate Studies Coordinator Limor Gindin<\n\nAttendance at the open day requires pre-registration.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Room 337 taub bld.
UID:eventx6a5a287eeb0ef10392
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20230419T190000
DTEND;TZID=Asia/Jerusalem:20230419T210000
DTSTAMP;TZID=Asia/Jerusalem:20230419T190000
SUMMARY: CSpecial Event  NVIDIA AI Lecture  at 2023-04-19 19:00:00
DESCRIPTION:You are invited to a joint lecture by the CS and EE on behalf of NVIDIA that will review AI and NVIDIA technologies, as well as the latest developments in the field of large language models, including effective tools for running large models, on Wednesday April 19, 2023 at 19:00, at the Faculty of Engineering Electrical, Auditorium 1003, Meyer Building.\n\nPlease  pre-register\n
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Room 1003, EE Meyer Building
UID:eventx6a5a287eeb0ff10399
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20230420T130000
DTEND;TZID=Asia/Jerusalem:20230420T140000
DTSTAMP;TZID=Asia/Jerusalem:20230420T130000
SUMMARY: CSpecial Talk  by Adam Kalai (Microsoft Research New England)  CS Lecture: The Power of Intelligent Language Models  at 2023-04-20 13:00:00
DESCRIPTION:Abstract: Recently, large language models have been trained on intelligent languages including natural languages, such as English, and programming languages, such as Python. We will examine several interesting applications of these models. First, they can be used to enumerate human stereotypes and discriminatory biases, suggesting that they must be used carefully. Second, they can be used to generate and solve their own programming puzzles, which can be used in a self-training pipeline to solve increasingly challenging algorithmic programming problems. Third, we illustrate how they can be used to simulate numerous human participants in classic behavioral economic and psychology experiments, such as the ultimatum game, risk aversion, garden path sentences, and the Milgram shock experiment. Finally, we discuss future directions in using these language models to understand intelligent animal communication in connection with Project CETI, which aims to understand the communication of sperm whales.\n\nBio: Adam Tauman Kalai is a Senior Principal Researcher at Microsoft Research New England. His research includes work on artificial intelligence and algorithms, with a focus on code generation and the responsible use of Language Models.  He received his BA from Harvard and PhD from at Carnegie Mellon University. He has served as an Assistant Professor at TTI-C and Georgia Tech. He is a member of the science team of Project CETI, an interdisciplinary initiative to understand the communication of sperm whales. He has co-chaired AI and crowdsourcing conferences including the COLT (the Conference on Learning Theory), the HCOMP (the Conference on Human Computation) and NEML. His honors include the Majulook prize, multiple best paper awards, an NSF CAREER award, and an Alfred P. Sloan fellowship.\n
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 601
UID:eventx6a5a287eeb10e10398
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20230423T143000
DTEND;TZID=Asia/Jerusalem:20230423T153000
DTSTAMP;TZID=Asia/Jerusalem:20230423T143000
SUMMARY: Coding_Theory_Semina  talk by Prof. Uzi Pereg  Coding Theory: The Multiple-Access Channel with Entangled Transmitters  at 2023-04-23 14:30:00
DESCRIPTION:Quantum communication has seen rapid development in the last decade, in both practice and theory. Recently, there is a growing interest in how quantum entanglement can assist classical networks, i.e., non-quantum communication systems. In particular, there are known examples of classical multi-user channels such that the sum rate with entangled transmitters is strictly higher than the best achievable sum rate without such resources. \n\nThe present work studies a two-user classical multiple-access channel (MAC) with entanglement resources shared between the transmitters a priori. We determine the capacity region for the general MAC with entangled transmitters, and show that the previous results can be obtained as a special case. We also point out the following change of behavior. Without entanglement resources, Dueck (1978) showed that the relaxation of a message-average error criterion can lead to strictly higher achievable rates, when compared with a maximal error criterion. Here, however, we show that the capacity region with entangled transmitters is the same, whether we consider a message-average or a maximal error criterion. \n\nUzi Pereg is an assistant professor at the Viterbi Faculty of Electrical and Computer Engineering (ECE) and the Hellen Diller Quantum Center in the Technion - Israel Institute of Technology. He was a postdoc at the Institute for Communications Engineering in the Technical University of Munich (TUM), and at the Munich Center for Quantum Science and Technology (2020-2022). He received his Ph.D. degree from the Technion in 2019. In September 2022, he joined the ECE faculty of the Technion. Uzi was awarded the Quantum Science and Technology Postdoc Fellowship of the Israel Council for Higher Education (CHE), the Seed Funding Grant of the Munich Center for Quantum Science and Technology (MCQST), the Chaya Career Advancement Chair, and the VATAT Fellowship for Junior Faculty Members in Quantum Science and Technology. \n
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 601
UID:eventx6a5a287eeb11e10397
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20230423T163000
DTEND;TZID=Asia/Jerusalem:20230423T173000
DTSTAMP;TZID=Asia/Jerusalem:20230423T163000
SUMMARY: PHD  talk by Sagi Marcovich  Balanced de Bruijn Sequences  at 2023-04-23 16:30:00
DESCRIPTION:Balanced sequences and balanced codes have attracted a lot of research in the last seventy years due to their diverse applications in information theory as well as other areas of computer science and engineering. There have been some methods to classify balanced sequences. This work suggests two new different hierarchies to classify these sequences. The first one is based on the largest $\ell$ for which each $\ell$-tuple is contained the same amount of times in the sequence. This property is a generalization for the property required for de Bruijn sequences. The second hierarchy is based on the number of balanced derivatives of the sequence. Enumeration for each such family of sequences and efficient encoding and decoding algorithms are provided in this work.\n
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture: 92840391109
UID:eventx6a5a287eeb13010383
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20230423T173000
DTEND;TZID=Asia/Jerusalem:20230423T193000
DTSTAMP;TZID=Asia/Jerusalem:20230423T173000
SUMMARY: CSpecial Event  Designated Meeting for Graduate Students: Research Career in Industry  at 2023-04-23 17:30:00
DESCRIPTION:You are invited to a designated meeting for graduate students, with a panel that will deal with research careers in industry: What does research in industry look like? What is the admission process for research positions? What are the types of jobs available and career paths? Featuring:\n\nDr. Liane Levy-Eitan, Research Group Director, Amazon\nAmichai Shulman, cyber entrepreneur and investor\nDr. Rachel Tzoref-Brill, Senior Researcher, IBM Research Laboratory\n\nThe meeting will take place on Sunday, May 23, 2023, 17:30, at the Grads Club in Taub (2nd floor).\n\nPlease pre-register.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:CS Grads Club, Floor 2, CS Taub Building
UID:eventx6a5a287eeb13f10396
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20230427T093000
DTEND;TZID=Asia/Jerusalem:20230427T103000
DTSTAMP;TZID=Asia/Jerusalem:20230427T093000
SUMMARY: MSC  talk by Dave Makhervaks  Clinical Contradiction Detection  at 2023-04-27 09:30:00
DESCRIPTION:Detecting contradictions in text is essential in determining the validity of the literature and sources that we consume. Medical corpora are riddled with conflicting statements. This is due to the large throughput of new studies and the difficulty in replicating experiments, such as clinical trials. Detecting contradictions in this domain is hard since it requires clinical expertise. In this work, we present a distant supervision approach that leverages a medical ontology to build a seed of potential clinical contradictions over 22 million medical abstracts. As a result, we automatically build a labeled training dataset consisting of paired clinical sentences that are grounded in an ontology and represent potential medical contradiction. The dataset is used to weakly-supervise state-of-the-art deep learning models showing significant empirical improvements across multiple medical contradiction datasets.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture: 99219466853 and Taub 601
UID:eventx6a5a287eeb14f10400
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20230502T103000
DTEND;TZID=Asia/Jerusalem:20230502T113000
DTSTAMP;TZID=Asia/Jerusalem:20230502T103000
SUMMARY: MSC  talk by Majd Khoury  On Distributed Computation of the Minimum Triangle Edge Transversal  at 2023-05-02 10:30:00
DESCRIPTION:In this work, we study the complexity of computing the distance of a graph from being triangle-free in distributed settings, that is, computing the minimum number of edges that must be removed to achieve a graph without triangles. We present lower bounds for the exact solution showing that this task is “as hard as it gets”. We also show fast algorithms for approximate solutions in multiple distributed models.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture: 96304259439  and Taub 601
UID:eventx6a5a287eeb15f10406
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20230502T113000
DTEND;TZID=Asia/Jerusalem:20230502T123000
DTSTAMP;TZID=Asia/Jerusalem:20230502T113000
SUMMARY: PHD  talk by Gregory Vaksman  Modern Learning Technics for Image and Video Denoising Via Patch Matching  at 2023-05-02 11:30:00
DESCRIPTION:Image and video denoising has been an area of research interest for decades. This talk will present three novel methods that take the denoising field a step forward. All the proposed methods in this work strongly rely on exploiting non-local self-similarity using patch matching. \n\nThe first method, termed LIDIA [1], has two contributions. First, we propose a low-weight architecture that achieves near state-of-the-art performance. Our architecture relies on patch matching and separable processing. Second, we introduce two simple and highly efficient methods for adapting the network to the input image, for boosting the denoising performance while addressing novel visual content that deviates from the training data. \n\nThe next method, termed PaCNet [2], is inspired by LIDIA, proposing a novel algorithm for video denoising. As in LIDIA, PaCNet relies on patch matching and separable processing. The proposed method uses the matched patches for constructing patch-craft frames, employing the latter as an augmentation of virtual frames for supporting the denoising task. Our algorithm achieves state-of-the-art performance, surpassing the competitors on average by about 0.7 dB. \n\nThe third method [3], inspired by both LIDIA and PaCNet, proposes a novel self-supervised training technique suitable for the removal of unknown correlated noise. The proposed approach neither requires knowledge of the noise model nor access to ground-truth targets. We assume that the noise is additive, zero mean, but not necessarily Gaussian, and one that could be short-range spatially correlated. Examples of such noise could be Gaussian correlated noise, shot noise passed through a linear space-invariant system, or real image noise in digital cameras. We demonstrate superior denoising performance compared to leading alternative self-supervised denoising methods.\n\n[1] G. Vaksman, M. Elad, and P. Milanfar. Lidia: Lightweight learned image denoising with instance adaptation. 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW), pages 2220-2229, 2020.\n[2] G. Vaksman, M. Elad, and P. Milanfar. Patch craft: Video denoising by deep modeling and patch matching. 2021 IEEE/CVF International Conference on Computer Vision (ICCV), pages 2137-2146, 2021.\n[3] G. Vaksman and M. Elad. Patch-Craft Self-Supervised Training for Correlated Image Denoising. To appear in the Conference on Computer Vision and Pattern Recognition (CVPR), 2023.\n
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 401
UID:eventx6a5a287eeb16d10403
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20230507T143000
DTEND;TZID=Asia/Jerusalem:20230507T153000
DTSTAMP;TZID=Asia/Jerusalem:20230507T143000
SUMMARY: MSC  talk by Avital Boruchovsky  DNA-Correcting Codes: End-to-end Correction in DNA Storage Systems  at 2023-05-07 14:30:00
DESCRIPTION:Existing storage technologies cannot keep up with the modern data explosion. There is a growing need to find alternatives for the current solutions for storing data. Storage systems based DNA, seems like an attractive possibility due to a number of unique properties of DNA mulecules, among them are that DNA is extremely dense (up to about 1  exabyte per cubic millimeter) and durable (half-life of over 500 years). \n\nA typical DNA storage system consists of three important components. The first is the DNA synthesis which produces the oligonucleotides, also called strands, that encode the data. The second part is a storage container with compartments which stores the DNA strands, however without order. Finally, to retrieve the data, the DNA is accessed using next-generation sequencing, which results in several noisy copies, called reads. The retrieval of the input information, is usually done by three steps as well. The first step is to partition all the reads into clusters such that the reads at each cluster are all copies of the same information strand. The second step is to apply a reconstruction algorithm on every cluster in order to retrieve an approximation of the original input strands. In the last step an Error Correcting Code is used in order to correct the remaining errors and to retrieve the user’s information.\n\nThis work presents a new solution to DNA storage that integrates all three steps of retrieval, namely clustering, reconstruction, and error correction. DNA-correcting codes are presented as a unique solution to the problem of ensuring that the output of the storage system is unique for any valid set of input strands. To this end, we introduce a novel distance metric to capture the unique behavior of the DNA storage system and provide necessary and sufficient conditions for DNA-correcting codes. The work also includes several upper bounds and constructions of DNA-correcting codes.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 601
UID:eventx6a5a287eeb18110404
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20230509T113000
DTEND;TZID=Asia/Jerusalem:20230509T123000
DTSTAMP;TZID=Asia/Jerusalem:20230509T113000
SUMMARY: PHD  talk by Tomer Weiss  Deep Learning Approaches for Inverse Problems in Computational Imaging and Chemistry  at 2023-05-09 11:30:00
DESCRIPTION:In this talk, I will present two chapters from my Ph.D. thesis. The core of my research focuses on methods that utilize the power of modern neural networks not only for their conventional tasks such as prediction or reconstruction, but rather use the information they “learned” (usually in the forms of their gradients) in order to optimize some end-task, draw insight from the data, or even guide a generative model.\n\nThe first part of the talk is dedicated to computational imaging and shows how to apply joint optimization of the forward and inverse models to improve the end performance. We demonstrate these methods on three different tasks in the fields of Magnetic Resonance Imaging (MRI) and Multiple Input Multiple Output (MIMO) radar imaging.\n\nIn the second part, we show a novel method for molecular inverse design that utilizes the power of neural networks in order to propose molecules with desired properties. We developed a guided diffusion model that uses the gradients of a pre-trained prediction model to guide a pre-trained unconditional diffusion model toward the desired properties. This method allows, in general, to transform any unconditional diffusion model into a conditional generative model.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture: 3369147024 and Taub 012
UID:eventx6a5a287eeb19310409
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20230509T143000
DTEND;TZID=Asia/Jerusalem:20230509T153000
DTSTAMP;TZID=Asia/Jerusalem:20230509T143000
SUMMARY: colloq  talk by Dan Halperin (Tel Aviv University)  CS Colloquia: From Snapping Fixtures to Multi-robot  Coordination: Geometry at the Service of Robotics  at 2023-05-09 14:30:00
DESCRIPTION:Robots sense, move and act in the physical world. It is therefore natural that understanding the geometry of the problem at hand is often key to devising an effective robotic solution. I will review several problems in robotics and automation in whose solution geometry plays a major role. These include designing optimized 3D printable fixtures, object rearrangement by robot arm manipulators, and efficient coordination of the motion of large teams of robots. As we shall see, exploiting geometric structure can, among other benefits, lead to reducing the dimensionality of the underlying search space and in turn to efficient solutions.\n\nShort bio:\nDan Halperin received his Ph.D. in Computer Science from Tel Aviv University, after which he spent three years at the Computer Science Robotics Laboratory at Stanford University. He then joined the Department of Computer Science at Tel Aviv University, where he is currently a full professor and for two years was the department chair. Halperin’s main field of research is Computational Geometry and Its Applications. Application areas he is interested in include robotics, automated manufacturing, algorithmic motion planning, and 3D printing. A major focus of Halperin’s work has been in research and development of robust geometric software, in collaboration with a group of European universities and research institutes: the CGAL project and library. Halperin was the program-committee chair/co-chair of several conferences in computational geometry, algorithms and robotics, including SoCG, WAFR, ESA, and ALENEX. Halperin is an ACM Fellow and an IEEE Fellow.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Room 337 taub bld.
UID:eventx6a5a287eeb1a310401
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20230509T180000
DTEND;TZID=Asia/Jerusalem:20230509T200000
DTSTAMP;TZID=Asia/Jerusalem:20230509T180000
SUMMARY: CSpecial Event  StarkWare Tech Talk  at 2023-05-09 18:00:00
DESCRIPTION:You are invited to a Tech Talk by StarkWare, a company that develops STARK-based solutions in the blockchain industry, on blockchain, the Scale problem, and zero-knowledge proofs (zk proofs), on Sunday, May 9, 2023, 18:00, at the Junta Bar, Technion.\n\nPlease pre-register.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Junta Bar, Technion
UID:eventx6a5a287eeb1c410405
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20230510T123000
DTEND;TZID=Asia/Jerusalem:20230510T143000
DTSTAMP;TZID=Asia/Jerusalem:20230510T123000
SUMMARY: CSpecial Event  Recruitment Day by Intuit  at 2023-05-10 12:30:00
DESCRIPTION:Intuit - a global fintech company - will hold a recruitment day and will present its business in Trust Data & Deep Insight, its technology and products, as well as vacancies, on Wednesday, May 10, 2023 starting at 12:30 in the Taub lobby.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:CS Taub Lobby
UID:eventx6a5a287eeb1d110408
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20230510T123000
DTEND;TZID=Asia/Jerusalem:20230510T133000
DTSTAMP;TZID=Asia/Jerusalem:20230510T123000
SUMMARY: Theory Seminar  talk by Amnon Ta-Shma (Tel-Aviv university)  Theory Seminar: HDX Condensers  at 2023-05-10 12:30:00
DESCRIPTION:More than twenty years ago, Capalbo, Reingold, Vadhan and Wigderson gave the first (and up to date only) explicit construction of a bipartite expander with almost full combinatorial expansion. The construction incorporates zig-zag ideas together with extractor technology, and is rather complicated. We give an alternative construction that builds upon recent constructions of hyper-regular, high-dimensional expanders. The new construction is, in our opinion, simple and elegant.\n\nBeyond demonstrating a new, surprising, and intriguing, application of high-dimensional expanders, the construction employs new ideas which we hope may lead to progress on the still remaining open problems in the area Joint work with Itay Cohen and Roy Roth from Tel-Aviv University.\n
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 201
UID:eventx6a5a287eeb1df10412
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20230510T183000
DTEND;TZID=Asia/Jerusalem:20230510T203000
DTSTAMP;TZID=Asia/Jerusalem:20230510T183000
SUMMARY: CSpecial Event  Practical Information Gathering Workshop by CYE  at 2023-05-10 18:30:00
DESCRIPTION:You are invited to an information gathering workshop CYE's Bug Bounty, led by Naftali Elazar, a cyber expert at CYE, and to hear about mapping the information gathering process as part of the process of identifying vulnerabilities in the organization, about tools and techniques for identifying organizational assets exposed to the Internet, and about gathering information using familiar tools and their use of the leading technologies in the market, on Wednesday, May 10, 2023, 18:30 at Taub 337.\n\nMore details and pre-registration - the number of places is limited, participation is subject to confirmation of registration.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Room 337 taub bld.
UID:eventx6a5a287eeb1ee10410
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20230514T113000
DTEND;TZID=Asia/Jerusalem:20230514T123000
DTSTAMP;TZID=Asia/Jerusalem:20230514T113000
SUMMARY: TDC  talk by Armando Castañeda (National Autonomous University of Mexico)  Distributed Computing Seminar: Asynchronous Wait-Free Runtime Verification and Enforcement of Linearizability  at 2023-05-14 11:30:00
DESCRIPTION:This work studies the problem of distributed runtime verification of linearizability for asynchronous concurrent implementations. It proposes an interactive model for distributed runtime verification and shows that it is impossible to verify at runtime this correctness condition for some common sequential objects such as queues, stacks, sets, priority queues, counters and the consensus problem. The impossibility captures informal arguments used in the past that argue distributed runtime verification is impossible.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 301
UID:eventx6a5a287eeb1fe10418
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20230514T143000
DTEND;TZID=Asia/Jerusalem:20230514T153000
DTSTAMP;TZID=Asia/Jerusalem:20230514T143000
SUMMARY: Coding_Theory_Semina  talk by Daniella Bar-Lev (CS, Technion)  Coding Theory: Cover Your Bases: How to Minimize the Sequencing Coverage in DNA Storage Systems  at 2023-05-14 14:30:00
DESCRIPTION:This seminar will be divided into two parts. In the first part, we will provide an introduction to DNA storage systems. This will include an overview of their biological and computational components, as well as a survey of the current technologies and emerging trends in the market landscape. \n\nIn the second part, we will focus on a novel problem called the DNA coverage depth problem. Motivated by the high cost and latency associated with DNA sequencing, we aim to design coding schemes that minimize the number of DNA strands that must be read to retrieve the desired information, while maintaining system reliability. Specifically, the DNA coverage depth problem seeks to optimize the required coverage depth as a function of the DNA storage channel, the error-correcting code, and the reconstruction algorithm. We will study the DNA coverage depth problem under both random and non-random access settings and explore coding schemes that optimize the required coverage depth.\n\nDaniella Bar-Lev is a Ph.D. student in the Computer Science Department at the Technion -- Israel Institute of Technology. She is a recipient of the Gutwirth Excellence Scholarship and the Student Research Prize for Cross-PI Collaboration in Data Science of VATAT. She received the B.Sc. degrees in computer science and mathematics, and an M.Sc. degree in computer science from the Technion -- Israel Institute of Technology, Haifa, Israel, in 2019 and 2021, respectively. Her research interests include algorithms, discrete mathematics, coding theory, and DNA storage\n
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 601
UID:eventx6a5a287eeb20c10415
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20230516T113000
DTEND;TZID=Asia/Jerusalem:20230516T123000
DTSTAMP;TZID=Asia/Jerusalem:20230516T113000
SUMMARY: pixel-club  talk by Niv Cohen (Hebrew University of Jerusalem)  Pixel Club: The Success and Challenges of Representation-Based Anomaly Detection  at 2023-05-16 11:30:00
DESCRIPTION:Anomaly detection aims to discover data which differ from the norm in a semantically meaningful manner. The task is difficult as anomalies are rare and unexpected. Moreover, a sample can be an important anomaly to one person and an uninteresting statistical outlier to another.\n\nIn this talk, I will first present how deep representations brought substantial gains for image anomaly detection and segmentation. Next, we will discuss the types of representations that are beneficial for anomaly detection, and how a given representation can be improved. Finally, I will present two remaining challenges and initial directions toward addressing them: (i) Strong nuisance variation, unrelated to the attributes we wish to inspect, may bias our representation (ii) Unexpected fine-grained combinations of normal parts (“logical anomalies”) may appear normal with coarse-grained representations.\n\nBio: Niv is a Ph.D. student at the Hebrew University of Jerusalem, advised by Dr. Yedid Hoshen. He received his BSc. in mathematics with physics, and M.Sc. in physics, both from the Technion. He's interested in computer vision and representation learning with a focus on anomaly detection and scientific data.\n
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Room 1061, EE Meyer Building
UID:eventx6a5a287eeb21d10416
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20230517T123000
DTEND;TZID=Asia/Jerusalem:20230517T133000
DTSTAMP;TZID=Asia/Jerusalem:20230517T123000
SUMMARY: colloq  talk by Thomas Vidick (Weizmann Institute of Science)  CS Colloquia: Testing Quantum Systems in the High-complexity Regime  at 2023-05-17 12:30:00
DESCRIPTION:From carefully crafted quantum algorithms to information-theoretic security in cryptography, a quantum computer can achieve impressive feats with no classical analogue. Can their correct realization be verified? When the power of the device greatly surpasses that of the user, computationally as well as cryptographically, what means of control remain available to the user? Recent lines of work in quantum cryptography and complexity develop approaches to this question based on the notion of an interactive proof. Generally speaking an interactive proof models any interaction whereby a powerful device aims to convince a restricted user of the validity of an agree-upon statement -- such as that the machine generates perfect random numbers or executes a specific quantum algorithm. Two models have emerged in which large-scale verification has been shown possible: either by placing reasonable computational assumptions on the quantum device, or by requiring that it consists of isolated components across which Bell tests can be performed.\n\nIn the talk I will discuss recent advances on the verification power of interactive proof systems between a quantum device and a classical user, focusing on the certification of quantum randomness from a single device, under cryptographic assumptions. \n\nBio:\n\nThomas Vidick is professor of Computer Science at the Weizmann Institute of Science, which he joined in 2022. Between 2014 and 2022 he was Assistant Professor, and then Professor, at the California Institute of Technology. Prior to joining Caltech, Vidick earned a B.A. in pure mathematics from Ecole Normale Superieure in Paris, a Masters in Computer Science from Universite Paris 7 and a Ph.D. from UC Berkeley. In 2020-2022 he was  a postdoctoral scholar at the Massachusetts Institute of Technology, supervised by Scott Aaronson.\n\nVidick's Ph.D. thesis was awarded the Bernard Friedman memorial prize in applied mathematics. In 2017 he was named a CIFAR Azrieli Global Scholar. In 2019 he received a Presidential Early-Career Award (PECASE). In 2021 he was named a Simons Investigator, and in 2023 he was awarded the Held prize from the US National Academy of Sciences. \n\nVidick's research is situated at the interface of theoretical computer science, quantum information and cryptography. He has investigated the role of entanglement in multiprover interactive proof systems and in quantum cryptography, making important contributions to both areas. 
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Room 337 taub bld.
UID:eventx6a5a287eeb23110402
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20230517T123000
DTEND;TZID=Asia/Jerusalem:20230517T143000
DTSTAMP;TZID=Asia/Jerusalem:20230517T123000
SUMMARY: CSpecial Event  Recruitment Day by Istra Research  at 2023-05-17 12:30:00
DESCRIPTION:Istra Research will hold a recruitment day and lecture on Wednesday, May 17, 2023, from 12:30-2:30 at the Taub Lobby, and at 13:00. there will be a lecture on Taub 3 (entrance floor) on &quot;Introduction to Algorithm Trading&quot; - High Frequency Trading - that you review basic concepts in the field.\n\nPlease register in advance \n\nRegistration for the Istra Riddle (announcement of the winner of the riddle and the prize will take place on May 31, 2023).
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:CS Taub Lobby and Taub 3
UID:eventx6a5a287eeb24610413
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20230517T173000
DTEND;TZID=Asia/Jerusalem:20230517T193000
DTSTAMP;TZID=Asia/Jerusalem:20230517T173000
SUMMARY: CSpecial Event  Round Tabels Event by Intel  at 2023-05-17 17:30:00
DESCRIPTION:You are invited to a round tables event with Intel researchers on the world of validation and how to deal with validation challenges using AI, on Tuesday, May 16, 2023, 17:30 in Taub 337.\n \nPlease pre-register.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Room 337 taub bld.
UID:eventx6a5a287eeb25610414
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20230522T130000
DTEND;TZID=Asia/Jerusalem:20230522T140000
DTSTAMP;TZID=Asia/Jerusalem:20230522T130000
SUMMARY: PHD  talk by Hadar Sivan  Efficient First and Second Order Methods for Function Monitoring and Optimization  at 2023-05-22 13:00:00
DESCRIPTION:Machine learning model training is a computationally expensive task that requires significant amounts of time and resources, especially for larger models. The problem is further increased when data arrives in a continuous stream since the model must be retrained multiple times to incorporate the new data and ensure the model remains accurate. Another difficulty arises during inference time when the data is geo-distributed; centralizing all data updates can be costly and lead to network overhead. To address these challenges, we propose various optimization and monitoring algorithms that integrate first and second-order information of the function to reduce optimization time and network overhead. Our proposed solutions aim to improve the efficiency and scalability of machine learning models, ultimately reducing the time and cost required for training and monitoring.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture: 97461530486 and Taub 601
UID:eventx6a5a287eeb26410407
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20230523T113000
DTEND;TZID=Asia/Jerusalem:20230523T123000
DTSTAMP;TZID=Asia/Jerusalem:20230523T113000
SUMMARY: pixel-club  talk by Gilad Lerman (University of Minnesota)  Pixel Club: Cycle-edge Message Passing for Group and Non-group Synchronization  at 2023-05-23 11:30:00
DESCRIPTION:The general synchronization problem asks to recover states of objects from their corrupted relative measurements. When the states are represented by group elements (e.g. 3-D rotations or permutations) this problem is known as group synchronization. In several applications, the algebraic structure of the states is more complicated, for example, the states can be represented by partial permutations. The synchronization problem has many applications, in particular, to structure-from-motion (SfM), where one needs to estimate the 3D structure of a scene from a set of its projected 2D images. I will first describe a general framework for group synchronization, the Cycle-Edge Message Passing (CEMP), and then explain its generalization to non-groups, by exemplifying the case of partial permutation synchronization. I will emphasize mathematical difficulties, review some mathematical guarantees for the proposed methods and also demonstrate an application. This is a joint work with Shaohan Li and Yunpeng Shi.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Room 1061, EE Meyer Building
UID:eventx6a5a287eeb27410422
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20230524T113000
DTEND;TZID=Asia/Jerusalem:20230524T123000
DTSTAMP;TZID=Asia/Jerusalem:20230524T113000
SUMMARY: ceClub  talk by Miron Livny (University of Wisconsin-Madison)  ceClub: Translational Computer Science at Work  at 2023-05-24 11:30:00
DESCRIPTION:The UW-Madison Center for High Throughput Computing (CHTC) is the home of the HTCondor Software Suite (HTCSS). Located in the Computer Sciences department, the center was established more than 15 years ago on the foundation of a research methodology that brings together innovation in distributed computing and services to scientists. Evaluation of new technologies under real-life conditions by engaged users advanced scientific discovery and guided the center in future research and development activities. The HTCSS has been serving for almost four decades as a means to place new capabilities in the hands of researchers who can benefit from advances in Throughput Computing. Having a capable and dependable suite of software tools enabled adoption of advanced Throughput Computing methodologies by an international community of users. Abramson and Parashar in their recent formalization of Translational Research in Computer Science (TCS) identified such an engaged community as one of the three pillars of the TCS workflow. These diverse adopters also provide the second pillar of the TCS workflow which is the locale – the place where the new technology is deployed and evaluated.\n\nThe talk will present the different elements that facilitate the translational work of the CHTC. These include a campus wide and a national research computing environment and a sustained sequence of software releases. The challenges of sustaining such a center in an academic institution will be discussed and the opportunities for innovation triggered by ever evolving research computing environments at all scales will be reviewed.\n\nBio:\nMiron Livny received a B.Sc. degree in Physics and Mathematics in 1975 from the Hebrew University and M.Sc. and Ph.D. degrees in Computer Science from the Weizmann Institute of Science in 1978 and 1984, respectively. Since 1983 he has been on the Computer Sciences Department faculty at the University of Wisconsin-Madison, where he is currently the John P. Morgridge Professor of Computer Science. He serves as the director of the Center for High Throughput Computing (CHTC), is leading the HTCondor Software Suite effort and serves as the technical director of the OSG. He is a member of the scientific leadership team of the Morgridge Institute of Research where he leads the Research Computing theme.\n\nDr. Livny’s research focuses on distributed processing and data management systems and involves close collaboration with researchers from a wide spectrum of disciplines. He pioneered the area of High Throughput Computing (HTC) and developed frameworks and software tools that have been widely adopted by academic and commercial organizations around the world.\n\nLivny is the recipient of the 2006 ACM SIGMOD Test of Time Award the 2013 HPDC Achievement Award and the 2020 IEEE TCDP Outstanding Technical Achievement Award.\n
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 401
UID:eventx6a5a287eeb28410423
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20230524T123000
DTEND;TZID=Asia/Jerusalem:20230524T133000
DTSTAMP;TZID=Asia/Jerusalem:20230524T123000
SUMMARY: Theory Seminar  talk by Uriya First (Haifa university)  Theory Seminar: A Sheaf-theoretic Approach to Constructing Locally Testable Codes  at 2023-05-24 12:30:00
DESCRIPTION:I will discuss a new approach towards constructing good locally testable codes (LTCs) with better qualities than the recent constructions of good LTCs. This approach continues the trend of using high dimensional expanders (HDXs) for constructing LTCs, but introduces a new ingredient: a sheaf on the HDX at hand. We show that if one could find a single example of a sheaved HDX satisfying some local expansion conditions and a cohomological condition --- both of which can be checked in finite (constant) time ---, then this example could be propagated into an infinite family of good LTCs. We also propose a heuristic method for constructing the initial sheaved HDX. The LTCs arising from our framework are 2-query LTCs which also admit a stronger testability property called (T)-testability.\n\nThis is a joint work with Tali Kaufman.\n
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 201
UID:eventx6a5a287eeb29910428
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20230524T130000
DTEND;TZID=Asia/Jerusalem:20230524T170000
DTSTAMP;TZID=Asia/Jerusalem:20230524T130000
SUMMARY: CSpecial Event  Excellence Program Alumni Conference  at 2023-05-24 13:00:00
DESCRIPTION:You are invited to celebrate 30 years of excellence: the Technion program for excellence is celebrating 30 years since its establishment in a series of lectures, on Wednesday, May 24, 2023 between 13:00-17:00, in Taub Auditorium 1, by program graduates and experts from academia and industry who will talk about their experience and insights on the trends, The latest innovations and challenges in their field.\n\nAmong the speakers: Prof. Ado Kaminer, Dr. Kira Radinsky, Prof. Nadav Cohen, Dr. Yair Wiener, Prof. Orr Dunkelman, Dr. Yaniv Altshuler, Limor Dori-Alon, Yoni Ackerman, Dr. Dean Leitersdorf and Orian Leitersdorf, and more.\n\nAfter the lectures, there will be a discussion on the subject of academia, industry and the defense system with the participation of Prof. Hagit Messer-Yeron, Aharon Aharon and Eyal Hulata, and moderator Chen Lieberman.\n\nDetails and full program.\n\nThe event is open to the public but requires pre-registration by Sunday, May 21, 2023.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:CS Taub Build. Auditorium 1
UID:eventx6a5a287eeb2a810425
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20230528T120000
DTEND;TZID=Asia/Jerusalem:20230528T130000
DTSTAMP;TZID=Asia/Jerusalem:20230528T120000
SUMMARY: MSC  talk by Natan Kaminsky  Lead Optimization For Drug Discovery With Limited Data  at 2023-05-28 12:00:00
DESCRIPTION:Drug development is a long and costly process consisting of several stages that can take many years to complete. \n\nOne of the early stage's goals is to optimize a novel chemical compound to be active against a target protein associated with the disease. \n\nThe goal of molecule optimization is, given an input molecule, to produce a new molecule that is chemically similar to the input molecule but with an improved property. \n\nIn this work, we present a novel approach for optimizing molecules. We propose to represent a molecule by breaking it into two disjoint substructures that we call: the molecule chains and the molecule core. \n\nWe train a model to generate the molecule chains with the desired properties for optimization, which are then attached to the molecule core to construct a novel molecule with high similarity to the input molecule.\n\nWe then show how to extend this approach to tasks where data is scarce, such as when attempting to target a drug to a novel protein. 
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture: 95830045630
UID:eventx6a5a287eeb2b810419
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20230528T143000
DTEND;TZID=Asia/Jerusalem:20230528T153000
DTSTAMP;TZID=Asia/Jerusalem:20230528T143000
SUMMARY: Coding_Theory_Semina  talk by Christoph Hofmeister (Technical University of Munich (TUM), Munich, Germany)  Coding Theory: How to Catch Liars in distributed Gradient Descent  at 2023-05-28 14:30:00
DESCRIPTION:This talk is about distributed machine learning in the presence of Byzantine errors. A main node performs gradient descent steps with the help of some worker nodes, a limited number of which are controlled by an adversary. These malicious worker nodes can return arbitrary data to the main node instead of the desired computation results. Prior work proposes distributing the data with redundancy among the workers and using error correction codes to detect and correct the erroneous computation results. In this work, we propose a solution that requires less redundancy at the cost of a small number of gradient computations on the main node and some light communication. In addition to a new scheme, we provide lower bounds on communication and computation.\n\nChristoph Hofmeister received a B.Eng. in electrical engineering and information technology from the Munich University of Applied Sciences (HM) in 2019 and an M.Sc. in electrical engineering and information technology from the Technical University of Munich (TUM) in 2021, where he is currently pursuing a Ph.D. with the Coding and Cryptography Group, Institute of Communications Engineering, under the supervision of Prof. Wachter-Zeh. His research interests include information and coding theory and its applications, with a focus on coded computing.\n
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 601
UID:eventx6a5a287eeb2c810429
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20230530T090000
DTEND;TZID=Asia/Jerusalem:20230530T150000
DTSTAMP;TZID=Asia/Jerusalem:20230530T090000
SUMMARY: CSpecial Event  Intel's Tech Experience  at 2023-05-30 09:00:00
DESCRIPTION:You are invited to Intel&#39;s Tech Experience event, on Tuesday, May 30, 2023, on the &quot;Shany&quot; Plaza of Taub Building:\nBetween 9:00-12:00 - AR/VR complex\nBetween 12:00-13:30 / 13:30-15:00 - two rounds of the FPGA workshop Hello world\n\nMore details, full program and pre-registration\n
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:"Shany" Plaza of Taub Building
UID:eventx6a5a287eeb2d910438
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20230531T123000
DTEND;TZID=Asia/Jerusalem:20230531T143000
DTSTAMP;TZID=Asia/Jerusalem:20230531T123000
SUMMARY: CSpecial Event  Recruitment Day by CYE  at 2023-05-31 12:30:00
DESCRIPTION:You are invited to recruitment day by CYE with engineers and recruitment teams and to a technological lecture by Dr. Nimrod Partosh, CS graduate and VP of AI at the company, which will deal with the question: How do you quantify the chance of a cyber attack in real organizations (and what do you do when the problem is NP-hard)? - on Wednesday, May 31, 2023 at 12:30 in the Taub lobby, and the lecture will take place at 13:30 in the Taub 012 Auditorium in the Learning Center on the entrance floor.\n\nPlease  pre-register.\n
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub  012 (Learning Center Auditorium)
UID:eventx6a5a287eeb2e710433
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20230531T123000
DTEND;TZID=Asia/Jerusalem:20230531T133000
DTSTAMP;TZID=Asia/Jerusalem:20230531T123000
SUMMARY: Theory Seminar  talk by Dean Doron (Ben-Gurion University)  Theory Seminar: Almost Chor–Goldreich Sources and Adversarial Random Walks  at 2023-05-31 12:30:00
DESCRIPTION:In this talk we consider the following adversarial, non-Markovian, random walk on “good enough” expanders: Starting from some fixed vertex, walk according to the instructions X = X1,…,Xt, where each Xi is only somewhat close to having only little entropy, conditioned on any prefix. The Xi-s are not independent, meaning that the distribution of the next step depends not only on the walk’s current node, but also on the path it took to get there.\n\nWe show that such walks (or certain variants of them) accumulate most of the entropy in X.\n\nWe call such X-s “almost Chor–Goldreich (CG) Sources”, and our result gives deterministic condensers with constant entropy gap for such sources, which were not known to exist even for standard CG sources, and even for the weaker model of Santha–Vazirani sources. \n\nAs a consequence, we can simulate any randomized algorithm with small failure probability using almost CG sources with no multiplicative slowdown. This result extends to randomized protocols as well, and any setting in which we cannot simply cycle over all seeds, and a “one-shot” simulation is needed.\n\nJoint work with Dana Moshkovitz, Justin Oh, and David Zuckerman\n
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 201
UID:eventx6a5a287eeb2f710436
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20230531T143000
DTEND;TZID=Asia/Jerusalem:20230531T153000
DTSTAMP;TZID=Asia/Jerusalem:20230531T143000
SUMMARY: PHD  talk by Idan Yaniv  Improving the Performance and Evaluation Methodology of Virtual Memory Systems  at 2023-05-31 14:30:00
DESCRIPTION:The virtual memory subsystem translates the address of each memory reference from its virtual to its physical representation, increasing execution runtimes by as much as 50% and 90% in bare-metal and virtual setups, respectively. We alleviate these overheads by developing improved virtual memory designs: (i) hashed page tables and (ii) TLB partitioning for simultaneous multithreading. We additionally develop an efficient and reliable methodology for evaluating the performance of newly proposed virtual memory designs, replacing conventional full-system, cycle-level simulations with much faster partial simulations of only the virtual memory subsystem, whose outputs are fed into a mathematical model that predicts the execution runtime.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture and Taub 601
UID:eventx6a5a287eeb30810424
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20230531T163000
DTEND;TZID=Asia/Jerusalem:20230531T173000
DTSTAMP;TZID=Asia/Jerusalem:20230531T163000
SUMMARY: MSC  talk by Boaz Moav  Tail-Erasure-Correcting Codes  at 2023-05-31 16:30:00
DESCRIPTION:The increasing demand for data storage has prompted the exploration of new techniques, with molecular data storage being a promising alternative. The stored information can be represented as a collection of two-dimensional arrays, such that each row represents a DNA strand. In this work, we present the results of our research into error-correcting codes for molecular data storage using this representation. Although both insertions and deletions have been observed to occur, the focus of our work is deletions, that can be caused by a failure of the bit addition chemistry. On top of this, cells can be lost either partially, which occurs when a defective bit prematurely terminates the chain, or the data within a cell can be corrupted completely. Initial experiments have reported error rates as high as 10%. Those errors can be considered as erasures in the last few symbols. Our study focuses on correcting those erasures and also deletions across rows. We present code constructions and explicit encoders that are shown to be nearly optimal in many scenarios, using bounds we derived. The first construction is based on permuting the columns of a chosen parity check matrix, such that the set of linearly independent columns match the pattern of possible erasures, that is at most a known parameter $d$ of tail-erasures. This construction is optimal for $d=2,3,4$ and nearly optimal for any $d=2t+1$. To design the above code properly, our work introduces a new distance metric, with properties similar to the Hamming distance, but which is better suited for the tail-erasure model. The second construction uses Tensor Product codes (TPC), that underlies Varshamov-Tenengolts (VT) codes to correct $t$ rows that suffer from one deletion each, where $t$ is a parameter. In the last construction we combine the two problems, to construct a code that can correct both tail-erasures and one deletion in $t$ rows. To conclude, as a concrete example, we mention Iridia, a San Diego-based startup, that has developed a new storage method using a two-dimensional arrays, for which our work suggests a suitable solutions. Our findings show that the new coding schemes are capable of effectively mitigating these errors, making Iridia's system a promising solution for DNA data storage.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 601
UID:eventx6a5a287eeb31810411
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20230531T190000
DTEND;TZID=Asia/Jerusalem:20230531T210000
DTSTAMP;TZID=Asia/Jerusalem:20230531T190000
SUMMARY: CSpecial Event  "Research on the Bar" Evening - TED Lectures  at 2023-05-31 19:00:00
DESCRIPTION:You are invited to the "Research on the Bar" evening - TED lectures and a meeting with three faculty members and their research groups on Wednesday, May 31, 2023 at 19:00 pm in Taub Terrace:\n\nProf. Eitan Yacobi: Storing information in DNA: Who ate my files?\nDr. Shaul Almagor: How to never make a mistake in anything\nDr. Ron Rothblum: How to prove without revealing anything\n\nPlease pre-register.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub Terrace
UID:eventx6a5a287eeb33610427
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20230601T110000
DTEND;TZID=Asia/Jerusalem:20230601T120000
DTSTAMP;TZID=Asia/Jerusalem:20230601T110000
SUMMARY: MSC  talk by  Yara Shamshoum  Measuring The Complexity of Neural Network Algorithms  at 2023-06-01 11:00:00
DESCRIPTION:Substantial efforts have been devoted into improving the capabilities of neural networks to solve algorithmic tasks. Through training, these networks learn to mimic algorithmic behaviour, enabling them to handle tasks such as sorting, navigating, and managing complex data structures like graphs. However, classic algorithms and neural networks are fundamentally different, making it challenging to analyze the complexity of an algorithm learned by a neural network. First, it is necessary to establish that the model has indeed learned an abstract solution resembling algorithmic behaviour. Additionally, while the complexity of classic algorithms is measured asymptotically, the complexity of an algorithm learned by a neural network must be measured within the finite input space supported by the model. This evaluation should also factor into consideration potential degradation in the model’s accuracy on unseen data. To this end, we will discuss these challenges and their implications, and propose methods for measuring the complexity of algorithms learned by neural networks. Finally, we will present experimental results on graph problems.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture: 93993434972  and Taub 601
UID:eventx6a5a287eeb34410439
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20230605T084500
DTEND;TZID=Asia/Jerusalem:20230607T210000
DTSTAMP;TZID=Asia/Jerusalem:20230605T084500
SUMMARY: CSpecial Event  SYSTOR 2023  at 2023-06-05 08:45:00
DESCRIPTION:You are invited to participate in the international conference SYSTOR 2023, leader in the fields of systems, cloud and storage, which will be held this year for the first time at the Technion, on Monday-Wednesday, June 5-7, 2023, at the Technion Samuel Neaman Institute for National Policy.\n\nParticipation is free of charge but requires pre-registration.\n\nMore details.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Technion Samuel Neaman Institute
UID:eventx6a5a287eeb36510420
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20230605T183000
DTEND;TZID=Asia/Jerusalem:20230605T203000
DTSTAMP;TZID=Asia/Jerusalem:20230605T183000
SUMMARY: CSpecial Event  NLP Research Session  at 2023-06-05 18:30:00
DESCRIPTION:You are invited to the research meeting - NLP Research Night (in collaboration with Grove Ventures - round tables with senior researchers from academia and industry in an open dialogue about the topics, challenges and burning questions in the field of NLP, on Monday, June 5, 2023, 18:30 in Taub Terrace.\n\nPre-registration is required, and is subject to approval.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub Terrace
UID:eventx6a5a287eeb37410432
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20230607T113000
DTEND;TZID=Asia/Jerusalem:20230607T123000
DTSTAMP;TZID=Asia/Jerusalem:20230607T113000
SUMMARY: ceClub  talk by Sam Noh (Virginia Tech)  ceClub: DyTIS: A Dynamic Dataset Targeted Index Structure Simultaneously Efficient for Search, Insert, and Scan  at 2023-06-07 11:30:00
DESCRIPTION:Many datasets in real life are complex and dynamic, that is, their key densities are varied over the whole key space and their key distributions change over time. It is challenging for an index structure to efficiently support all key operations for data management, in particular, search, insert, and scan, for such dynamic datasets. In this talk, I will present DyTIS (Dynamic dataset Targeted Index Structure), an index that targets dynamic datasets. DyTIS, though based on the structure of Extendible hashing, leverages the CDF of the key distribution of a dataset, and learns and adjusts its structure as the dataset grows. The key novelty behind DyTIS is to group keys by the natural key order and maintain keys in sorted order in each bucket to support scan operations within a hash index. We also define what we refer to as a dynamic dataset and propose a means to quantify its dynamic characteristics. Our experimental results show that DyTIS provides higher performance than the state-of-the-art learned index for the dynamic datasets considered.\n\nBio:\nSam H.(Hyuk) Noh received the BE degree in computer engineering from the Seoul National University, Seoul, Korea, in 1986, and the PhD degree from the Department of Computer Science, University of Maryland, College Park, MD, in 1993. He held a visiting faculty position at the George Washington University, Washington, DC, from 1993 to 1994 before joining Hongik University, in Seoul, Korea, where he was a professor in the School of Computer and Information Engineering until the Spring of 2015. During this period, he served as the Chair of the Department of Computer Engineering as well as the Head of the School from September of 2013 through February of 2015. From August 2001 to August 2002, he was also a visiting associate professor with the University of Maryland Institute of Advanced Computer Studies (UMIACS), College Park, MD. Starting from the Fall of 2015 he joined UNIST (Ulsan National Institute of Science and Technology), a young science and tech focused national university, where he was a Professor at the Department of Computer Science and Engineering and served as the inaugural Dean of the Graduate School of Artificial Intelligence in the College of Information and Biotechnology from August 2020 through March 2023. He also served as the Dean of the School of Electrical and Computer Engineering from January of 2016 through June of 2018. As of January 2023, he is a Professor at the Computer Science Department at Virginia Tech. He has served/serves as General Chair, Program Chair, and Program Committee Member on a number of technical conferences and workshops. He also served on the Steering Committee of LCTES from 2016 through 2020. He is currently the Chair of the Steering Committee for ACM HotStorage and a Steering Committee member of USENIX FAST and IEEE NVMSA. He also served as Editor-in-Chief of the ACM Transactions on Storage from 2016 through 2022. His research interests include system software issues pertaining to computer systems in general and storage systems in particular, with a focus on the use of new memory technologies such as flash memory and persistent memory. He is a Fellow of the ACM and IEEE and a member of USENIX and KIISE (Korean Institute of Information Scientists and Engineers).\n
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 401
UID:eventx6a5a287eeb38710441
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20230607T123000
DTEND;TZID=Asia/Jerusalem:20230607T133000
DTSTAMP;TZID=Asia/Jerusalem:20230607T123000
SUMMARY: MSC  talk by Adi Amuzig  Value of Assistance for Mobile Agents  at 2023-06-07 12:30:00
DESCRIPTION:Mobile robotic agents often suffer from localization uncertainty which grows with time and with the agents' movement. This can hinder their ability to accomplish their task. In some settings, it may be possible to perform assistive actions that reduce uncertainty about a robot’s location. Since assistance may be costly and limited, and may be requested by different members of a team, there is a need for principled ways to support the decision of which assistance to provide to an agent and when, as well as to decide which agent to help within a team. For this purpose, we propose Value of Assistance (VOA) to represent the expected cost reduction that assistance will yield at a given point of execution.  We offer a way to compute VOA based on an estimation of the robot's future uncertainty, modeled as a Gaussian process. We specify conditions under which our VOA measure is valid, and empirically demonstrate the ability of our measure to predict the agent's average cost reduction when receiving assistance in both simulated and real-world robotic settings. 
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture: 92471871959
UID:eventx6a5a287eeb3a210421
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20230607T123000
DTEND;TZID=Asia/Jerusalem:20230607T133000
DTSTAMP;TZID=Asia/Jerusalem:20230607T123000
SUMMARY: Theory Seminar  talk by Roy Gotlib (Bar-Ilan University)  Theory Seminar: List Agreement Testing and High Dimensional Expansion  at 2023-06-07 12:30:00
DESCRIPTION:One of the key components in PCP constructions are agreement tests.\n\nIn agreement testing the tester is given access to subsets of fixed size of some set, each equipped with an assignment.\n\nThe tester is then tasked with testing whether these local assignments agree with some global assignment over the entire set.\n\nOne natural generalization of this concept is the case where, instead of a single assignment to each local view, the tester is given access to $\ell$ different assignments for every subset.\n\nThe tester is then tasked with testing whether there exist $\ell$ global functions that agree with all of the assignments of all of the local views.\n\nIn this talk I will present a recent result that shows that if the subsets form a (sufficiently good) high dimensional expander then they support list agreement testing under mild assumptions on the local assignments.\n\nIn addition, I will also show that those mild assumptions are necessary for list agreement.\n\nI will not assume any prior knowledge of high dimensional expanders.\n\nBased on a joint work with Tali Kaufman\n
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 201
UID:eventx6a5a287eeb3b210443
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20230607T153000
DTEND;TZID=Asia/Jerusalem:20230607T163000
DTSTAMP;TZID=Asia/Jerusalem:20230607T153000
SUMMARY: ceClub  talk by Yuli Mandelblat (Intel)  ceClub: Recent Intel Hybrid CPU Architecture  at 2023-06-07 15:30:00
DESCRIPTION:Why hybrid and what problems this technology solves?\nWhat is the difference between Intel’s Hybrid Technology and the other market solutions (e.g. Big-little)?\nHybrid micro-architectural solutions: caches, fabric.\nHow SW knows what core to use for what task.\nIntel Thread Director - what is does and why it is required.\nFuture development of Hybrid solutions.\n\nBio:\nYuli (Julius) Mandelblat is an Intel Fellow of Client SoC Architecture Team. \nYuli works at Intel since 1990. Throughout his career, he worked on multiple Intel CPUs in variety of areas of product development from validation and design to architecture. He was responsible for the development of the first multi-core solutions, on-die interconnect, memory ordering and coherency solutions, etc. In the latest Intel® client products Yuli led the definition of highly successful Hybrid architecture that was a key for the success of CPU’s of 12th and 13th generations of Intel CPU.\nYuli holds M.Sc. degree from Russian University of Transport (MIIT).\nYuli is an inventor of more than 20 patents in different areas of CPU architecture.\n
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Room 352, EE Meyer Building
UID:eventx6a5a287eeb3c310442
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20230607T163000
DTEND;TZID=Asia/Jerusalem:20230607T173000
DTSTAMP;TZID=Asia/Jerusalem:20230607T163000
SUMMARY: PHD  talk by Maria Abu Sini  The Reconstruction Model and Shortmers-based DNA Synthesis  at 2023-06-07 16:30:00
DESCRIPTION:Levenshtein's reconstruction model was first introduced in 2001 and suggests transmitting a word over multiple noisy channels, then using the channels' outputs to recover the transmitted word. This talk will discuss the reconstruction model when the channels are prone to combinations of errors, or when unique retrieval of the transmitted word is not guaranteed to succeed. In particular, when the channels introduce a limited number of insertions (or deletions), and unique decoding is not guaranteed to succeed, bounds on the largest list size will be presented. Furthermore, a recently proposed optimization to DNA synthesis using shortmers (i.e., sequences of bases) will be investigated from a theoretical point of view. This optimization differs from the conventional synthesis process by appending in each cycle not only single bases, but also shortmers. The significance of this optimization lies in reducing the number of cycles, which then determines the time and monetary cost of the synthesis process. Hence, the talk will discuss several questions pertaining to this optimization, such as which shortmers to use and how to calculate the minimum number of cycles.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 601
UID:eventx6a5a287eeb3d310430
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20230608T110000
DTEND;TZID=Asia/Jerusalem:20230608T120000
DTSTAMP;TZID=Asia/Jerusalem:20230608T110000
SUMMARY: colloq  talk by Prof. Bjarne Stroustrup  CS Colloquia: C++20 – Reaching for the Aims of C++  at 2023-06-08 11:00:00
DESCRIPTION:Out of necessity C++ has been an evolving language. I outline some early ideals for C++, some techniques for keeping the evolution directed, and show how C++20 comes close to many of those ideals. Specific topics include  type-and-resource safe code, generic programming, modularity,  the elimination of the preprocessor, and error handling. Naturally, over the years, C++ has acquired many “barnacles” that can become obstacles to developing elegant and efficient code. That has been a recognized problem since the early days of C – Dennis Ritchie and I talked about it – so we must distinguish between what can be done and what should be done. The C++ Core Guidelines is the current best effort in that direction.\n\nShort bio:\nBjarne Stroustrup is the designer and original implementer of C++ as well as the author of The C++ Programming Language (4th Edition) and A Tour of C++ (2nd edition), Programming: Principles and Practice using C++ (2nd Edition), and many popular and academic publications.\n\nDr. Stroustrup is a Technical Fellow and Managing Director in the technology division of Morgan Stanley in New York City as well as a visiting professor at Columbia University. He is a member of the US National Academy of Engineering, and an IEEE, ACM, and CHM fellow. He is the recipient of the 2018 NAE Charles Stark Draper Prize for Engineering and the 2017 IET Faraday Medal. He did much of his most important work in Bell Labs. His research interests include distributed systems, design, programming techniques, software development tools, and programming languages. To make C++ a stable and up-to-date base for real-world software development, he has been a leading figure with the ISO C++ standards effort for 30 years.\n\nHe holds a master’s in Mathematics from Aarhus University and a PhD in Computer Science from Cambridge University, where he is an honorary fellow of Churchill College.\n\nSpace is limited - Please register in advance
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub TBD
UID:eventx6a5a287eeb3e310440
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20230611T143000
DTEND;TZID=Asia/Jerusalem:20230611T153000
DTSTAMP;TZID=Asia/Jerusalem:20230611T143000
SUMMARY: Coding_Theory_Semina  talk by Prof. Ohad Elishco (Ben-Gurion University)  Coding Theory: Codes Over Absorption Channels  at 2023-06-11 14:30:00
DESCRIPTION:In recent years, extensive research has been dedicated to the development of nano- and micro-machines. While the majority of practical research is focused on chemistry and biology, there is also research aimed at communication aspects. This is crucial because nano-machines are limited in their capabilities and require communication and networking to tackle complex tasks. By collaborating, these machines can revolutionize medicine by serving as intelligent drug delivery systems, advanced sensors, and more.  \n\nThe primary challenge in nano-machine communication lies in their inability to utilize electromagnetic waves for communication. Hence, an alternative communication method must be employed. When operating in living organisms (in-vivo), one potential approach is to leverage the nervous system, which serves as nature’s communication medium.  \n\nIn this presentation, we introduce a novel communication channel called the absorption channel, inspired by information transmission through neurons. Our motivation stems from the potential applications of in-vivo nano- and micro-machines, advancements in medical technology, and brain-machine interfaces that communicate via the nervous system.  \n\nWe will commence by providing a motivation for the proposed channel. Subsequently, we will present codes capable of correcting absorption errors for any given finite alphabet. We will explore various scenarios, including single absorption error correction over binary alphabets, as well as correction codes for general alphabets. If time permits, we will also delve into multiple absorption error-correcting codes over general alphabets. \n\nBio:\nOhad Elishco received his B.Sc., M.Sc, and Ph.D degrees in electrical engineering from Ben-Gurion University of the Negev, Israel. Between 2017-2018 he was a postdoc at the RLE at MIT, hosted by Prof. Muriel Medard. Between 2018-2020 he was a postdoc at ISR at UMD, hosted by Prof. Alexander Barg. Since 2020 he has been an assistant professor at Ben-Gurion University of the Negev. His research interests include constrained coding, Information and coding for biology, and dynamical systems.\n
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 601
UID:eventx6a5a287eeb3f710444
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20230612T113000
DTEND;TZID=Asia/Jerusalem:20230612T123000
DTSTAMP;TZID=Asia/Jerusalem:20230612T113000
SUMMARY: MSC  talk by Guy Horowitz  Causal Strategic Classification  at 2023-06-12 11:30:00
DESCRIPTION:When users can benefit from certain predictive outcomes, they may be prone to act to achieve those outcome, e.g., by strategically modifying their features. The goal in strategic classification is therefore to train predictive models that are robust to such behavior. However, the conventional framework assumes that changing features does not change actual outcomes, which depicts users as "gaming" the system. Here we remove this assumption, and study learning in a causal strategic setting where true outcomes do change. Focusing on accuracy as our primary objective, we show how strategic behavior and causal effects underlie two complementing forms of distribution shift. We characterize these shifts, and propose a learning algorithm that balances between these two forces and over time, and permits end-to-end training. Experiments on synthetic and semi-synthetic data demonstrate the utility of our approach.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 601
UID:eventx6a5a287eeb40a10437
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20230612T143000
DTEND;TZID=Asia/Jerusalem:20230612T153000
DTSTAMP;TZID=Asia/Jerusalem:20230612T143000
SUMMARY: colloq  talk by Avi Wigderson (IAS Princeton)  CS Special Guest Lecture: The Value of Errors in Proofs  at 2023-06-12 14:30:00
DESCRIPTION:CS Special Guest Lecture by Prof. Avi Wigderson, IAS Princeton, on the Occasion of his Being Awarded an Honorary Doctorate from the Technion\n\nRecently, a group of theoretical computer scientists posted a paper on the Arxiv with the strange-looking title "MIP* = RE", surprising and impacting not only complexity theory but also some areas of math and physics. Specifically, it resolved, in the negative, the "Connes' embedding conjecture" in the area of von-Neumann algebras, and the "Tsirelson problem" in quantum information theory. It further connects Turing's seminal 1936 paper which defined algorithms to Einstein's 1935 paper with Podolsky and Rosen which challenged quantum mechanics.\n\nAs it happens, both acronyms MIP* and RE represent proof systems, of a very different nature. To explain them, we'll take a meandering journey through the classical and modern definitions of proof. I hope to explain how the methodology of computational complexity theory, especially modelling and classification (of both problems and proofs) by algorithmic efficiency, naturally leads to the generation of new such notions and results (and more acronyms, like NP). A special focus will be on notions of proof which allow interaction, randomness, and errors, and their surprising power and magical properties.\n\nThe talk does not require special mathematical background.\n\n
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Room 337 taub bld.
UID:eventx6a5a287eeb41810426
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20230613T143000
DTEND;TZID=Asia/Jerusalem:20230613T153000
DTSTAMP;TZID=Asia/Jerusalem:20230613T143000
SUMMARY: colloq  talk by Matan Gavish (The Hebrew University of Jerusalem)  CS Colloquia: Power, Responsibility and Computer Science Training  at 2023-06-13 14:30:00
DESCRIPTION:Over the course of four decades, the academic field of computer science transformed from a branch of mathematics to a key driver of the evolution of our species.  Graduates of academic computer science programs today routinely create systems that would have been considered, just a century ago, miracles of mythic proportions.  Judging by cultural impact, computer science departments today resemble Hogwarts much more than they resemble the theoretical havens they used to be before computers got seriously strong.  \n\nInterestingly, at their core, computer science BSc programs have not changed that much since the 1990's. If indeed with great power comes great responsibility, then it is now incumbent on academic computer science departments to prepare the young witches and wizards, who train in CS, to use their digital powers responsibly - whatever this may mean. Needless to say, almost nothing in the academic computer science literature offers any clues on how to go about this.  On the contrary, computer science - like all science and technology - detaches technique from its human impact, focuses almost exclusively on problem solving, and tends to view the world through a narrow quantitative lens. Questions of human impact are typically labelled under the obscure term "ethics" and deferred wholesale to the social sciences and humanities.  Clear basic guidelines for individual choices, analogous to those of the medical and life sciences, still seem lightyears ahead.\n\nI will argue that academic computer science departments offer the perfect environment to be asking these questions - as Joseph Weizenbaum and Norbert Wiener passionately prophesied at the onset of the digital age - and that it may be our solemn duty to do so.  I'll share some experiences from teaching an undergraduate CS class in Hebrew University, which looked for meaningful ways to form a personal perspective on the broader implications of digital information technology.\n \nShort Bio: \n \nMatan Gavish is an Associate Professor in the Hebrew University School of Computer Science and Engineering. He is the founder of the Israel-Singapore center for AI-based urban agriculture (iSURF), and of the Hebrew University joint CS-Statistics BSc program in Data Science. His research interests include statistical learning, mathematical statistics of spectral algorithms in high dimensions, applied harmonic analysis, random matrix theory, empirical mathematics, reproducible research, data-driven precision agriculture, digital communication studies, and philosophy of technology. Matan received the dual B.Sc. degree in Mathematics and Physics from Tel Aviv University in 2006, the M.Sc. degree in Mathematics from the Hebrew University of Jerusalem in 2008 and the Ph.D. degree in Statistics from Stanford University in 2014. 
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Room 337 taub bld.
UID:eventx6a5a287eeb42910435
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20230613T160000
DTEND;TZID=Asia/Jerusalem:20230613T170000
DTSTAMP;TZID=Asia/Jerusalem:20230613T160000
SUMMARY: MSC  talk by Dan Navon  A Robust Approach to Vision-Based Terrain Aided Localization  at 2023-06-13 16:00:00
DESCRIPTION:Terrain-aided navigation (TAN) was developed before the GPS era to prevent the error growth of inertial navigation. TAN algorithms were initially developed to exploit altitude over ground or clearance measurements from a radar altimeter in combination with a Digital Terrain Map (DTM). After almost two decades of silence, the availability of inexpensive cameras and computational power and the need to find efficient GPS-denied positioning solutions have prompted a renewed interest in this solution. However, vision-based TAN is more challenging in many aspects than the original one, as visual observables can only provide a range up to a scale, preventing a straightforward extension of classical TAN techniques.\n\nThe main contributions of this work are the introduction of a new, more flexible, and efficient algorithm for solving the visual-assisted TAN. The algorithm combines two fast stages for solving the problem. In addition, a new outlier-rejection step is introduced between the two stages to make the algorithm robust and suitable for real-world data.\n\n
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture: 94171574353 and Taub 601.
UID:eventx6a5a287eeb44410431
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20230614T123000
DTEND;TZID=Asia/Jerusalem:20230614T133000
DTSTAMP;TZID=Asia/Jerusalem:20230614T123000
SUMMARY: Theory Seminar  talk by Michal Dory (Haifa university)  Theory Seminar: Approximate All-Pairs Shortest Paths: Recent Advances and Open Questions  at 2023-06-14 12:30:00
DESCRIPTION:The All-Pairs Shortest Paths (APSP) problem is one of the most fundamental problems in graph algorithms. It is well-known that APSP can be solved in O(n^3) time in weighted graphs, and in O(n^{omega}) time in unweighted graphs, where omega
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 201
UID:eventx6a5a287eeb45910447
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20230614T163000
DTEND;TZID=Asia/Jerusalem:20230614T173000
DTSTAMP;TZID=Asia/Jerusalem:20230614T163000
SUMMARY: MSC  talk by Dganit Hanania  On the Capacity of DNA Labeling  at 2023-06-14 16:30:00
DESCRIPTION:DNA labeling is a powerful tool in molecular biology and biotechnology that allows for the visualization, detection, and study of DNA at the molecular level. Under this paradigm, a DNA molecule is being labeled by specific k patterns and is then imaged. Then, the resulted image is modeled as a (k + 1)-ary sequence in which any non-zero symbol indicates on the appearance of the corresponding label in the DNA molecule. The primary goal of this work is to study the labeling capacity, which is defined as the maximal information rate that can be obtained using this labeling process. The labeling capacity is computed for any single label and several results are provided for multiple labels as well. Moreover, we provide the optimal minimal number of labels of length one or two that are needed in order to gain labeling capacity of 2.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 601
UID:eventx6a5a287eeb46d10417
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20230618T143000
DTEND;TZID=Asia/Jerusalem:20230618T153000
DTSTAMP;TZID=Asia/Jerusalem:20230618T143000
SUMMARY: Coding_Theory_Semina  talk by Yonatan Yehezkeally (University of Munich)  Coding Theory: Resilient Repeat-free Codes  at 2023-06-18 14:30:00
DESCRIPTION:Repeat-free codes are used to ensure unique reconstruction from fragmentation, assuming full (uniform) read-coverage of substrings, with applications to DNA-based storage systems. In this talk, we explore a generalization aimed at resilience to pre-fragmentation noise, and study existence results as well as explicit constructions.\n\nYonatan Yehezkeally is the Carl Friedrich von Siemens post-doctoral research fellow of the Alexander von Humboldt Foundation, in the Associate Professorship of Coding and Cryptography (Prof. Wachter-Zeh), School of Computation, Information and Technology, Technical University of Munich. His research interests include coding for novel storage media, with a focus on DNA-based storage and nascent sequencing technologies, as well as combinatorial structures and finite group theory.\n\nYonatan received the Ph.D. degree in Electrical and Computer Engineering in 2020, from Ben-Gurion University of the Negev, Beer-Sheva, Israel. Before that, he received the B.Sc. degree (cum laude) in Mathematics and the M.Sc. degree (summa cum laude) in Electrical and Computer Engineering, in 2013 and 2017 respectively, also from Ben-Gurion University of the Negev.\n
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 601
UID:eventx6a5a287eeb47e10448
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20230620T143000
DTEND;TZID=Asia/Jerusalem:20230620T153000
DTSTAMP;TZID=Asia/Jerusalem:20230620T143000
SUMMARY: colloq  talk by Prof. Adi Shamir (Weizmann Institute of Science)  CS Colloquia: Facial Misrecognition Systems  at 2023-06-20 14:30:00
DESCRIPTION:In this talk I will describe how to plant novel types of backdoors in any facial recognition model based on the popular architecture of deep Siamese neural networks, by mathematically changing a small fraction of its weights (i.e., without using any additional training or optimization). These backdoors force the system to err only on specific persons which are preselected by the attacker. For example, we show how such a backdoored system can take any two images of a particular person and decide that they represent different persons (an anonymity attack), or take any two images of a particular pair of persons and decide that they represent the same person (a confusion attack), with almost no effect on the correctness of its decisions for other persons. Uniquely, we show that multiple backdoors can be independently installed by multiple attackers who may not be aware of each other's existence with almost no interference. \nJoint work with Irad Zehavi.Bio:Adi Shamir is a Professor of Computer Science at the Weizmann Institute of Science. Professor Shamir's research focuses on the foundations and applications of cryptography. Among his research contributions are the RSA algorithm, the SSS scheme and differential cryptanalysis. He is the recipient of various awards and honors for his scientific contributions, including the Turing award, Erdos prize, and the Israel Prize.\n\n
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Room 337 taub bld.
UID:eventx6a5a287eeb49110445
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20230621T123000
DTEND;TZID=Asia/Jerusalem:20230621T133000
DTSTAMP;TZID=Asia/Jerusalem:20230621T123000
SUMMARY: Theory Seminar  talk by Noam Mazor (Tel-Aviv university)  Theory Seminar: Incompressiblity and Next-Block Pseudoentropy  at 2023-06-21 12:30:00
DESCRIPTION:A distribution is k-incompressible, Yao [FOCS ’82], if no efficient compression scheme compresses it to less than k bits. While being a natural measure, its relation to other computational analogs of entropy such as pseudoentropy, Hastad, Impagliazzo, Levin, and Luby [SICOMP 99], and to other cryptographic hardness assumptions, was unclear.\n\nWe advance towards a better understating of this notion, showing that a k-incompressible distribution has (k-2) bits of next-block pseudoentropy, a refinement of pseudoentropy introduced by Haitner, Reingold, and Vadhan [SICOMP ’13]. We deduce that a samplable distribution X that is (H(X) + 2)-incompressible, implies the existence of one-way functions.\n\nJoint work with Iftach Haitner and Jad Silbak.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 201
UID:eventx6a5a287eeb4a810452
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20230621T153000
DTEND;TZID=Asia/Jerusalem:20230621T163000
DTSTAMP;TZID=Asia/Jerusalem:20230621T153000
SUMMARY: ceClub  talk by Omri Palmon (Storage Architecture for HPC)  ceClub: “Accelerated CPU Computing” Course  at 2023-06-21 15:30:00
DESCRIPTION:In this talk, we will review the various needs and solutions for storage in HPC environments. We will compare the requirements for storage when used for data input and output, scratch space, or inter-server communication, and review the various solutions for them, When reviewing solutions, we will analyze the various access protocols, technologies and specific solutions. We will also compare on-prem storage solutions to cloud-based ones,\n\nBio:\nDr. Omri Palmon has a Ph.D. in computer science from Stanford University, and a BSc. and MSc. degrees in pure mathematics from Tel Aviv University. Dr. Palmon is currently one of the co-founders of Weka.IO, a leader in modern scalable file system development, and was formerly VP product for XIV, which developed SAN solutions and was acquired by IBM.\n
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Room 352, EE meyer Building
UID:eventx6a5a287eeb4bb10453
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20230622T100000
DTEND;TZID=Asia/Jerusalem:20230622T110000
DTSTAMP;TZID=Asia/Jerusalem:20230622T100000
SUMMARY: CSpecial Talk  by Leshem Choshen (IBM)  CS Guest Lecture: Collaborative Pretraining and Recycling Finetuned Models to do So  at 2023-06-22 10:00:00
DESCRIPTION:This talk will discuss our recent advancements in recycling finetuned models and collaborative pretraining. We would describe how to harness the data and computation invested in one or more models to collaboratively improve the pre-trained model they originated from, once or over and over again. The work will also touch on our initial understanding of how and why fusing several models by weight averaging works. All of these are small steps towards evolving pretrained models that we create together as a community, join us, check the best models, and feel free to contact me and ask questions.\n\nBio:\nLeshem Choshen currently leads the ColD-fusion challenge at IBM, aiming to collaboratively pretrain and propose to recycle finetuned models to do so. He received the postdoctoral Rothschild and Fulbright fellowship as well as IAAI and Blavatnik best Ph.D. awards. With broad NLP and ML interests, he also worked on Reinforcement Learning, Evaluation and Understanding of how neural networks learn. In parallel, he participated in Project Debater, creating a machine that could hold a formal debate, ending in a Nature cover and live debate.\n\nHe is also a dancer and runs a food and science blog (NisuiVeTeima on Instagram, Facebook and Tiktok).\n
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 601
UID:eventx6a5a287eeb4cf10455
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20230622T153000
DTEND;TZID=Asia/Jerusalem:20230622T163000
DTSTAMP;TZID=Asia/Jerusalem:20230622T153000
SUMMARY: MSC  talk by Aviram Imber  Inference over Elections with Incomplete Information  at 2023-06-22 15:30:00
DESCRIPTION:A central task in social choice is that of aggregating voter preferences to decide who wins. For this task, a voting rule maps a collection of voter preferences over the candidates to a set of winning candidates. Relevant scenarios may be political elections, document rankings in search engines, hiring dynamics in the job market, and so on. We study situations in which voter preferences are incomplete. These scenarios arise naturally in a variety of practical settings: voters may be undecided about some candidates, new candidates can be introduced, and the information can be retrieved from indirect sources such as social media.We analyze the complexity of some fundamental computational problems in such settings. In our first work, we study the task of computing the minimal and maximal ranks that a candidate can obtain, given partial voting preferences. In our second work, we study the problem of computing the probability of winning in an election where voter attendance is random. In our third work, we introduce several models of incomplete votes for approval-based committee (ABC) voting. We study the problems of determining whether a given set of candidates is a possible or necessary winning committee. In our fourth work, we consider spatial voting, where candidates and voters are located in the Euclidean space, and each voter ranks candidates based on their distance from the voter's position. We study the problems of finding the possible and necessary winners given incomplete information about the voter's position.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture: 3525715262  and Taub 401
UID:eventx6a5a287eeb4e410450
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20230627T113000
DTEND;TZID=Asia/Jerusalem:20230627T123000
DTSTAMP;TZID=Asia/Jerusalem:20230627T113000
SUMMARY: cggc  talk by Prof. Alla Sheffer (University of British Columbia)  CGGC Seminar: Human-Centered Geometry Processing  at 2023-06-27 11:30:00
DESCRIPTION:Humans can ubiquitously communicate and reason about both tangible and abstract shape properties. Artists can succinctly convey complex shapes to a broad audience using a range of mediums; and human observers can effortlessly analyze and agree on observed shape properties such as upright-orientation or style. While perception research provides some clues as to the mental processes humans employ when performing these tasks, concrete and quantifiable explanations of these actions are frequently lacking. Our recent research aims to quantify the geometric properties underlying human shape communication and analysis, and to develop algorithms that successfully replicate human abilities in these domains. In my talk I will survey our efforts in this space, focusing on ways to incorporate insights about human perception into  algorithm design.  My talk will include examples across a wide range of 2D and 3D geometry processing tasks, including shape orientation, VR interfaces for shape modeling, raw sketch consolidation;  clip-art vectorization; clip-art reshaping;  sketch-based 3D reconstruction; and style analysis and transfer for man-made shapes. The common thread in our proposed solutions to these problems is the use of insights derived from perception and design literature combined with derivation of quantitative properties via targeted human perception studies and  machine learning from scarce data.\n\nBio:\nAlla Sheffer received her PhD from the Hebrew University in 1999 and is currently a full professor at the University of British Columbia, Canada, where she investigates algorithms for shape modeling and analysis in the context of computer graphics applications. She is best known for her research on mesh parameterization, hexahedral meshing, computational garment design, and perception driven shape modeling. Dr. Sheffer is a Fellow of ACM, Fellow of IEEE and Fellow of the Royal Society of Canada. She is a Member of SIGGRAPH Academy,  a  recipient of the Canadian Human Computer Communications Society Achievement Award’18 and a UBC Killam Research Award'19. Her research has been supported by faculty awards from IBM, Google and Adobe, a Killam Research Fellowship, and an Audi Production Award. Dr. Sheffer has served as an Associate Editor of all three major computer graphics journals (ACM Transactions on Graphics, IEEE Transactions on Visualization Computer Graphics, and Eurographics Computer Graphics Forum). She is the Technical Papers Chair for SIGGRAPH'23 and served as a program co-chair for Eurographics’18, Symposium on Geometry Processing’06, and Shape Modeling’13. She was a general co-chair for the Pacific Graphics’18 and Geometric Modeling and Processing’19 conferences. Dr. Sheffer had co-authored over 100 peer-reviewed publications, including 50 papers in ACM Transactions on Graphics, the topmost competitive CG venue. She holds six recent patents on methods for sketch analysis and hexahedral mesh generation.\n
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub  012 (Learning Center Auditorium)
UID:eventx6a5a287eeb4f910449
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20230627T113000
DTEND;TZID=Asia/Jerusalem:20230627T123000
DTSTAMP;TZID=Asia/Jerusalem:20230627T113000
SUMMARY: pixel-club  talk by Yoav Berlatzky (PxE Holographic Imaging)  Pixel Club: How to Beat classical Light-field Imaging’s Resolution Llimit  at 2023-06-27 11:30:00
DESCRIPTION:Light-field cameras promised to revolutionize imaging by capturing and recording the propagation paths of light-rays through space. This light-field information, equivalent to canonical optical phase space, supposedly holds the “sys-admin” password to optical imaging. Digital post-processing and manipulation can allow digital refocusing of rays, correction of optical aberrations, as well as calculating the distance to every point in the imaged scene. However, once this technology was put into practice, its severe limitations and trade-offs became apparent. The main problem is that conventional light-field imaging contains an inherent trade-off between the light-rays’ angular information and the overall image resolution, analogous to Heisenberg’s uncertainty principle. PxE Holographic Imaging has developed a white-light holographic imaging camera that overcomes these limitations allowing highly accurate depth inference, digital refocusing and deblurring, as well as holographic, wavefront, and spectral imaging to be performed with no compromise on image resolution. In this talk we’ll explain the physics behind this breakthrough, and it’s relation to the optical coherence matrix formalism, the Wigner distribution, von Neumann measurements, and positive-operator-valued measures (POVMs).\n\nYoav Berlatzky is CEO and co-founder of PxE Holographic Imaging. Prior to PxE, Yoav held engineering and managerial positions at Applied Materials and Nova Measuring Instruments. He holds a Ph.D. in physics from the Technion and has over 40 patents granted and pending.\n
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Room 1061, EE Meyer Building
UID:eventx6a5a287eeb50e10451
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20230627T123000
DTEND;TZID=Asia/Jerusalem:20230627T143000
DTSTAMP;TZID=Asia/Jerusalem:20230627T123000
SUMMARY: CSpecial Event  Projects Fair on IoT, Android, Arduino and Networks  at 2023-06-27 12:30:00
DESCRIPTION:You are invited to the CS Projects Fair for the Spring Semester of 2023, where 30 teams of undergraduate students will present and demonstrate projects in various fields in IoT, Android, Arduino and Networks, developed as part of the final project in the software engineering and communication networks track, most of which were carried out in collaboration with various social associations and organizations, and were intended to make a contribution to the community.\n\nThe fair will be held on Tuesday, June 27, 2023, 12:30-14:30, at the CS Taub Lobby.\n\nThe presenting posters (Heb)
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:CS Taub Lobby
UID:eventx6a5a287eeb52510456
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20230628T113000
DTEND;TZID=Asia/Jerusalem:20230628T123000
DTSTAMP;TZID=Asia/Jerusalem:20230628T113000
SUMMARY: cggc  talk by Prof. Amir Vaxman (University of Edinburgh)  CGGC Seminar: Unconventional Fields: What are these Vectors Actually Good For  at 2023-06-28 11:30:00
DESCRIPTION:Directional and vector fields are central objects in geometry processing. They are commonly represented with low-order simple elements on watertight surfaces in FEM and in computer graphics, for simplicity and sparsity. Moreover, they are classically defined only on conventional symmetries. I will discuss some recent works of extending these classical representations with alternative bases, where then important tasks in downstream applications like meshing, computations of flows, making Penrose patterns on surfaces, or properly marketing mangoes can be done robustly and more efficiently.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub  012 (Learning Center Auditorium)
UID:eventx6a5a287eeb53810434
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20230628T113000
DTEND;TZID=Asia/Jerusalem:20230628T123000
DTSTAMP;TZID=Asia/Jerusalem:20230628T113000
SUMMARY: ceClub  talk by Yuval Yarom (Ruhr University Bochum) - CANCELLED!  ceClub: Just About Time  at 2023-06-28 11:30:00
DESCRIPTION:When multiple programs execute on the same computer, they share the use of the microarchitectural resources. Because program execution affects the state of the microarchitecture and the state of the microarchitecture affects program execution time, measuring execution time can reveal information on the state of the microarchitecture, and with it on prior execution of other programs. Thus, such micoroarchitectural timing attacks leak information by measuring variations in program execution time. \n\nAs these attacks often measure minute timing variations, at the order of few nanoseconds, multiple proposed defences aim at depriving attackers of high-resolution clocks. In response, counter-proposals that show how to overcome these defences have been published. In this talk we look at the ensuing armed-race and explore techniques for limiting timer resolution and for carrying out attacks with restricted timers. We will take a close look at the impact of low-resolution clocks on microarchitectural attacks, explore techniques for amplifying signals by over six orders of magnitude, and demonstrate how attackers can perform high-frequency, high-resolution attacks without using high-resolution clocks.\n\nBio:\nProfessor Yuval Yarom holds the chair for Computer Security at Ruhr University Bochum. He earned his Ph.D. in Computer Science from the University of Adelaide in 2014, and an M.Sc. in Computer Science and a B.Sc. in Mathematics and Computer Science from the Hebrew University of Jerusalem in 1993 and 1990, respectively. In between he has been the Vice President of Research at Memco Software and a co-founder and Chief Technology Officer of Girafa.com.\n\nYuval's research explores the security of the interface between the software and the hardware. In particular, He is interested in the discrepancy between the way that programmers think about software execution and the concrete execution in modern processors. He works on identifying micro-architectural vulnerabilities, and on exploitation and mitigation techniques.\n
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Room 815, EE Meyer Building
UID:eventx6a5a287eeb54910458
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20230628T123000
DTEND;TZID=Asia/Jerusalem:20230628T143000
DTSTAMP;TZID=Asia/Jerusalem:20230628T123000
SUMMARY: CSpecial Event  Best Project Contest - The Finals  at 2023-06-28 12:30:00
DESCRIPTION:You are invited to the Finals event of the Best Project Competition, which will be held in the format of a project fair, and everyone is invited to encourage the competing teams and watch the most creative projects.\n\nThe event will take place on Wednesday, June 28, 2023 on the entrance floor of the Taub Computer Science Building:\n12:30 - Project fair in the CS lobby of the Taub Building - entrance floor\n14:00 - Announcement of the winners and awarding of certificates and prizes in Taub 2 Auditorium- entrance floor\n\n
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:CS Taub Lobby and Taub Auditorium 2
UID:eventx6a5a287eeb55c10457
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20230628T123000
DTEND;TZID=Asia/Jerusalem:20230628T133000
DTSTAMP;TZID=Asia/Jerusalem:20230628T123000
SUMMARY: Theory Seminar  talk by Yuval Rabani (The Hebrew University of Jerusalem)  Theory Seminar: The Randomized $k$-Server Conjecture is False!  at 2023-06-28 12:30:00
DESCRIPTION:We prove a few new lower bounds on the randomized competitive ratio for the $k$-server problem and other related problems, resolving some long-standing conjectures. In particular, for metrical task systems (MTS) we asympotically settle the competitive ratio and obtain the first improvement to an existential lower bound since the introduction of the model 35 years ago (in 1987). \nMore concretely, we show: \n\n1. There exist $(k+1)$-point metric spaces in which the randomized competitive ratio for the $k$-server problem is $\Omega(\log^2 k)$. This refutes the folklore conjecture (which is known to hold in some families of metrics) that in all metric spaces with at least $k+1$ points, the competitive ratio is $\Theta(\log k)$. \n2. Consequently, there exist $n$-point metric spaces in which the randomized competitive ratio for MTS is $\Omega(\log^2 n)$. This matches the upper bound that holds for all metrics. The previously best existential lower bound was $\Omega(\log n)$ (which was known to be tight for some families of metrics). \n3. For all $k < n \in ℕ$, for *all* $n$-point metric spaces the randomized $$k-server competitive ratio is at least $\Omega(\log k)$, and consequently the randomized MTS competitive ratio is at least $\Omega(\log n)$. These universal lower bounds are asymptotically tight. The previous bounds were $\Omega(\log k / \log\log k)$ and $\Omega(\log n / \log\log n)$, respectively. \n4. The randomized competitive ratio for the $w$-set metrical service systems problem, and its equivalent width-$w$ layered graph traversal problem, is $\Omega(w^2)$. This slightly improves the previous lower bound and matches the recently discovered upper bound. \n5. Our results imply improved lower bounds for other problems like $k$-taxi, distributed paging and metric allocation. \n\nThese lower bounds share a common thread, and other than the third bound, also a common construction.\n\nThis is joint work with Sebastien Bubeck and Christian Coester.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 201
UID:eventx6a5a287eeb56c10459
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20230703T120000
DTEND;TZID=Asia/Jerusalem:20230703T130000
DTSTAMP;TZID=Asia/Jerusalem:20230703T120000
SUMMARY: MSC  talk by Daniel Gilo  A General Search-based Framework for Generating Textual  at 2023-07-03 12:00:00
DESCRIPTION:One of the prominent methods for explaining the decision of a machine-learning classifier is by a counterfactual example. Most current algorithms for generating such examples in the textual domain are based on generative language models.  Generative models, however, are trained to minimize a specific loss function in order to fulfill certain requirements for the generated texts.  Any change in the requirements may necessitate costly retraining, thus potentially limiting their applicability. We present a general search-based framework for generating counterfactual explanations in the textual domain.  Our framework is model-agnostic,  domain-agnostic, anytime, and is able to adapt to changes in the user requirements without retraining. We model the task as a search problem in a space where the initial state is the classified text, and the goal state is a text in a given target class.  Our framework includes domain-independent modification operators, but can also exploit domain-specific knowledge through specialized operators. The search algorithm attempts to find a text from the target class with minimal user-specified distance from the original classified object. 
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture: 92212370744  and Taub 401
UID:eventx6a5a287eeb57f10454
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20230704T133000
DTEND;TZID=Asia/Jerusalem:20230704T143000
DTSTAMP;TZID=Asia/Jerusalem:20230704T133000
SUMMARY: PHD  talk by Niv Giladi  Enabling Scalable Learning with Large Models  at 2023-07-04 13:30:00
DESCRIPTION:Deep Neural Networks (DNNs) training continues to scale over size and computational footprint, as a result of a higher number of trainable parameters, wider and deeper models, and growing amounts of training data. As improvements in model quality lead over hardware capabilities, this scale-up translates into a need for a growing number of training devices working in tandem, turning distributed training into the standard approach for training DNNs on a large scale. This seminar delves into distributed training, exploring the current solutions and implications for improving scalability. First, we will examine asynchronous training from a dynamical stability perspective, and derive optimal hyperparameters tuning rules. Then, we will look into scalability challenges in synchronous training and suggest a method to improve its robustness. Finally, we will introduce a paradigm of integrating deep learning with physics simulations to improve the scalability of the latter, leading to x4096 theoretical acceleration in physics simulation.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 601
UID:eventx6a5a287eeb5ac10460
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20230705T113000
DTEND;TZID=Asia/Jerusalem:20230705T123000
DTSTAMP;TZID=Asia/Jerusalem:20230705T113000
SUMMARY: ceClub  talk by Kfir Girstein (EE, Technion)  ceClub: Cyber Attack Simulation Infrastructure for Effective Detection in Real-Time Systems  at 2023-07-05 11:30:00
DESCRIPTION:Real-time systems are designed to respond to external stimuli and complete tasks within a predetermined timeframe. The development of these systems often involves the use of cycle-accurate simulation environments and digital twin systems to accurately model the system and its operating environment. Ensuring high reliability and security in real-time systems is essential, and the development environment must incorporate events related to reliability, such as sensor failure and subsystem malfunction, as well as cyber security attacks. This thesis proposes a cyber attack simulation infrastructure that extends simulation environments of real-time systems and incorporates events related to reliability and advanced cyber security attacks, including attacks on single and multiple sensors. The proposed infrastructure also enables the use of simulation capabilities to train data-based algorithms, such as machine learning, which is difficult to achieve on live systems. The proposed environment is validated using a drone’s flight control system and navigation system, which utilizes machine learning algorithms. To support the proposed environment, the SCART framework and dataset are introduced, which efficiently increase model accuracy and significantly reduce false-positive rates.  Some of these experiments were also validated using a set of ” real drones“.\n\nKfir is an M.Sc. student under the supervision of Prof. Avi Mendelson.\n
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Room 861, EE Meyer Building & Zoom Lecture: 94673013539
UID:eventx6a5a287eeb5bd10464
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20230705T123000
DTEND;TZID=Asia/Jerusalem:20230705T133000
DTSTAMP;TZID=Asia/Jerusalem:20230705T123000
SUMMARY: Theory Seminar  talk by Jonathan Mosheiff (Ben-Gurion university)  Theory Seminar: On Continuous Analogues of LDPCs and LTCs  at 2023-07-05 12:30:00
DESCRIPTION:A critical notion in coding theory is that of a&nbsp;good&nbsp;code—a&nbsp;code&nbsp;with constant rate and distance.&nbsp;A natural analogous notion in the p-norm&nbsp;(1 &lt;= p &lt;= 2)&nbsp;is that of a&nbsp;good&nbsp;l_p-spread subspace. A linear subspace C \subset R^n is called&nbsp;good&nbsp;l_p-spread&nbsp;if&nbsp;dim(C) &gt;= Omega(n) &quot;constant rate&quot;), and every&nbsp;x \in C \ {0} is at least&nbsp;Omega(|x|_p)-far (in&nbsp;l_p-distance)&nbsp; from any&nbsp;O(n)-sparse vector. The p=2&nbsp;case, in which&nbsp;C&nbsp;is called a&nbsp;Euclidean Section,&nbsp;is of particular importance.\n\nIn this talk we focus on&nbsp;sparse&nbsp;l_p-spread matrices, namely, sparse matrices whose kernel is a good&nbsp;l_p-spread subspace. Such a matrix is analogous to the parity-check matrix of an&nbsp;LDPC&nbsp;code. We also study the stronger notion of a&nbsp;sparse&nbsp;(Omega(n), O(1))-RIP (restricted isometry property) matrix, which we consider as an analogue to an&nbsp;LTC (Locally testable&nbsp;code).\n\nIn this regime, we analyze the l_p-spread and l_p-RIP for random sparse matrices. We prove that sparse l_p-RIP matrices exist for all 1 &lt;= p &lt; 2, but do not exist for p=2. We also explicitly construct&nbsp;l_p-RIP matrices for&nbsp;1&lt;= p &lt;= p_0, where&nbsp;p_0 &gt; 1 is a universal constant.&nbsp;&nbsp;Based on joint works with Venkat Guruswami and Peter Manohar.\n
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 201
UID:eventx6a5a287eeb5ce10463
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20230709T110000
DTEND;TZID=Asia/Jerusalem:20230709T120000
DTSTAMP;TZID=Asia/Jerusalem:20230709T110000
SUMMARY: PHD  talk by Alona Levy    Deep Learning and Statistical Methods for Digital Pathology and Molecular Measurements  at 2023-07-09 11:00:00
DESCRIPTION:Digital analysis of pathology whole-slide images is fast becoming a game changer in cancer diagnosis and treatment. Specifically, deep learning methods have shown great potential to support pathology analysis, with recent studies identifying molecular traits that were not previously recognized in pathology H&E whole-slide images. Simultaneous to these developments, it is becoming increasingly evident that tumor heterogeneity is an important determinant of cancer prognosis and susceptibility to treatment, and should therefore play a role in the evolving practices of matching treatment protocols to patients. In this talk, I will present our work on spatially resolving bulk mRNA and miRNA expression levels on pathology whole-slide images (WSIs). I will further present a statistical method we developed to spatially characterize tumor heterogeneity from the inferred gene expression levels and demonstrate that it applies to a wide variety of spatial data, including 3D data like brain MRI scans. Finally, I will present a deep learning model we developed to predict DNA methylation levels.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture: 93993434972  and Taub 601
UID:eventx6a5a287eeb5e010462
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20230711T113000
DTEND;TZID=Asia/Jerusalem:20230711T123000
DTSTAMP;TZID=Asia/Jerusalem:20230711T113000
SUMMARY: PHD  talk by Bahjat Kawar  Diffusion Models for Image Restoration  at 2023-07-11 11:30:00
DESCRIPTION:Denoising Diffusion Probabilistic Models (DDPM), also known as diffusion models, have recently emerged as state-of-the-art generative models, synthesizing images with unprecedented quality and realism. At their core, diffusion models employ an MSE-trained denoiser neural network in an iterative scheme, transforming random noise into pristine images. Theoretically, this algorithm is proven to draw samples from a learned prior image distribution. In our work, we adapt pre-trained diffusion models for the task of image restoration, mainly focusing on linear inverse problems. We present a novel outlook on inverse problem solving, posing it as a posterior sampling task rather than an optimization problem. This approach introduces several advantages such as improved perceptual quality, multiple solutions, and uncertainty quantification. Moreover, our method does not require task-specific training, and we demonstrate its use for inpainting, super resolution, deblurring, colorization, and more.\n
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture: 96270781265  and Taub 401
UID:eventx6a5a287eeb5f210446
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20230716T100000
DTEND;TZID=Asia/Jerusalem:20230716T110000
DTSTAMP;TZID=Asia/Jerusalem:20230716T100000
SUMMARY: MSC  talk by  Liran Farhi  Movement as a Language: Unleashing the Power of Indoor Movement Analysis for Semantic Place Prediction  at 2023-07-16 10:00:00
DESCRIPTION:The proliferation of modern mobile phones has opened up unprecedented opportunities for leveraging location-tracking capabilities to extract individual mobility patterns and contextual information. However, existing approaches heavily rely on analyzing geolocation data obtained from Global Navigation Satellite System (GNSS) observations, which are limited when used in enclosed spaces. This talk presents new algorithms for efficient mobility data analysis and contextual learning by utilizing the WiFi infrastructure that exists today in most indoor environments. Firstly, I will introduce SWATSON, an unsupervised trajectory segmentation algorithm that is designed to partition a continuous sequence of data points into homogeneous segments. SWATSON effectively utilizes temporal constraints, mitigates noise, and exhibits applicability across various domains. By employing SWATSON, I will demonstrate an approach to analyzing movement in indoor environments without needing a localization process. Consequently, the integration of SWATSON will be explored to create personalized semantic categorical place labeling, such as assigning the label "work" to specific locations. I will present a supervised learning-based classification model, leveraging WiFi-based attributes for sequence-based modeling and spatio-temporal feature generation to enhance the accuracy of semantic place labeling. Experiments show a 97% V-measure score in segmenting data point into distinct spaces, notably surpassing the performance of GNSS-based solutions in semantic place labeling, thereby highlighting substantial progress in the fields of trajectory analysis and predictive modeling.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 601
UID:eventx6a5a287eeb60410465
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20230723T133000
DTEND;TZID=Asia/Jerusalem:20230723T143000
DTSTAMP;TZID=Asia/Jerusalem:20230723T133000
SUMMARY: ceClub  talk by Joseph Friedman (University of Texas at Dallas)  ceClub: Reversible, Neuromorphic, Reservoir, and Secure Computing with Spintronic Phenomena  at 2023-07-23 13:30:00
DESCRIPTION:The rich physics present in a wide range of spintronic materials and devices provide opportunities for a variety of computing applications. This presentation will describe six distinct proposals to leverage spintronic phenomena for reversible computing, neuromorphic computing, reservoir computing, and hardware security. The presentation will begin with a solution for reversible computing in which magnetic skyrmions propagate and interact in a scalable system with the potential for energy dissipation below the Landauer limit, followed by a paradigm for operating Boolean logic at terahertz clock frequencies utilizing the magnetoresistance of low-dimensional materials. Three neuromorphic systems for emulating neurobiological behavior with spintronic phenomena will then be presented: a purely-spintronic system that enables unsupervised learning with magnetic domain wall neurons and synapses, a reservoir computing system based on the dynamics of frustrated nanomagnets, and an approach for unsupervised learning that marks the first experimental demonstration of a neuromorphic network directly implemented with MTJ synapses. This presentation will conclude with a logic locking paradigm based on nanomagnet logic, the first logic locking system that is secure against both physical and algorithmic attacks.\n\nBio:\nDr. Joseph S. Friedman is an associate professor of Electrical & Computer Engineering at The University of Texas at Dallas and director of the NeuroSpinCompute Laboratory. He holds a Ph.D. and M.S. in Electrical & Computer Engineering from Northwestern University and undergraduate degrees from Dartmouth College. He was previously a CNRS Research Associate with Université Paris-Saclay, a Summer Faculty Fellow at the U.S. Air Force Research Laboratory, a Visiting Professor at Politecnico di Torino, a Guest Scientist at RWTH Aachen University, and worked on logic design automation at Intel Corporation.\n\nDr. Friedman is a member of the editorial boards of Scientific Reports and IEEE Transactions on Nanotechnology, and previously the Microelectronics Journal. He is a conference chair of SPIE Spintronics, has served on numerous conference technical program committees, and is the founder and chairperson of the Texas Symposium on Computing with Emerging Technologies (ComET). He has also been awarded the National Science Foundation (NSF) Faculty Early Career Development Program (CAREER) Award.\n
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Room 861, EE Meyer Building & Zoom Lecture: 94673013539
UID:eventx6a5a287eeb61710466
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20230725T113000
DTEND;TZID=Asia/Jerusalem:20230725T123000
DTSTAMP;TZID=Asia/Jerusalem:20230725T113000
SUMMARY: PHD  talk by Roee Francos  Multi-Agent Teamwork in Search for Smart Opponents Detection  at 2023-07-25 11:30:00
DESCRIPTION:Cooperative Multi-Agent teams can be deployed in many interesting and important domains such as industry, transportation, agriculture, security and more. In this talk, I will introduce key results from my research, primarily focusing on theoretical work concerned with search for smart agents by UAV teams. Suppose that in a given planar circular region, there are some smart mobile agents, and we would like to find them using teams of sweeping agents. A smart agent is an agent capable of detecting and responding to the motions of searchers by performing evasive maneuvers, to avoid detection. We assume various search configurations for the sweeping team of agents, and present guaranteed search techniques for single agent and multi agent teams. These search procedures enable both confinement of the smart agents to their original domain as well as complete detection of all of them by searching the entire expanding domain. Furthermore, we investigate the dual problem of devising guaranteed defense policies for protecting a given region from the entrance of smart mobile agents by detecting them using a team of sweeping agents. The desired outcome of the developed protocols is a defense strategy of the original domain and for its expansion.Afterwards, I will briefly discuss two other avenues of research I am investigating. The first is a biodynamical analysis of locust trajectories towards the purpose of understanding and modeling movements of locust swarms.  The objective of this research is to assist in developing bio-inspired robotic technology by learning from insect modes of locomotion and enabling the biological and robotics communities to collaboratively work toward bio-inspired locomotion. The second is concerned with developing algorithms for intelligent transportation systems aimed at enabling provably safe and efficient management and routing of large numbers of aerial vehicles for Urban Air Mobility applications.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 401
UID:eventx6a5a287eeb62d10467
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20230731T103000
DTEND;TZID=Asia/Jerusalem:20230731T123000
DTSTAMP;TZID=Asia/Jerusalem:20230731T103000
SUMMARY: CSpecial Event  Maternitech Community Meeting - A Community For Technological Parents On Maternity Leave From The "Faculty of Computer Science at the Technion"  at 2023-07-31 10:30:00
DESCRIPTION:&nbsp; To participate, register at the linkIn the program:Dr. Inbal Tsafir-Lavia, CEO and co-entrepreneur at Nevia Bio, which is developing a test for the early diagnosis of ovarian cancerMichal Shanhav, Talent Acquisition Partner, IBM Research, HR IsraelTo the community page on Facebook&nbsp;
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:CS Taub Building
UID:eventx6a5a287eeb64210478
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20230807T123000
DTEND;TZID=Asia/Jerusalem:20230807T133000
DTSTAMP;TZID=Asia/Jerusalem:20230807T123000
SUMMARY: MSC  talk by  Dan Aaronson  PB-FS: Postcard Based Fast Start  at 2023-08-07 12:30:00
DESCRIPTION:We propose PB-FS (Postcard-Based Fast Start), a rate initialization scheme that uses direct feedback from the switches to quickly correct the rates of new datacenter flows that begin at the line rate and cause congestion. PB-FS is designed to easily integrate into any datacenter congestion control protocol. We evaluate PB-FS in two datacenter environments: a lossless network that runs RoCE, and a lossy network that uses RDMA with selective repeat. We show that PB-FS significantly reduces tail latency of short flows while maintaining throughput, in both lossless and lossy datacenters.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture: 95365724708
UID:eventx6a5a287eeb65410479
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20230808T160000
DTEND;TZID=Asia/Jerusalem:20230808T170000
DTSTAMP;TZID=Asia/Jerusalem:20230808T160000
SUMMARY: MSC  talk by Tomer Cohen  Improved Approximation for Two-dimensional Vector Multiple Knapsack  at 2023-08-08 16:00:00
DESCRIPTION:We study the uniform 2-dimensional vector multiple knapsack (2VMK) problem, a natural variant of multiple knapsack arising in real-world applications such as virtual machine placement. The input for 2VMK is a set of items, each associated with a 2-dimensional weight vector and a positive profit, along with m 2-dimensional bins of uniform (unit) capacity in each dimension. The goal is to find an assignment of a subset of the items to the bins, such that the total weight of items assigned to a single bin is at most one in each dimension, and the total profit is maximized.Our main result is a (1−ln2−ε)-approximation algorithm for 2VMK, for every fixed ε>0, thus improving the best known ratio of (1−1/e−ε) which follows as a special case from a result of [Fleischer at al., MOR 2011]. Our algorithm relies on an adaptation of the Round&Approx framework of [Bansal et al., SICOMP 2010], originally designed for set covering problems, to maximization problems. The algorithm uses randomized rounding of a configuration-LP solution to assign items to ≈m⋅ln2≈0.693⋅m of the bins, followed by a reduction to the (1-dimensional) Multiple Knapsack problem for assigning items to the remaining bins.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture: 94011474168  and Taub 601
UID:eventx6a5a287eeb66510468
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20230809T110000
DTEND;TZID=Asia/Jerusalem:20230809T120000
DTSTAMP;TZID=Asia/Jerusalem:20230809T110000
SUMMARY: MSC  talk by Tomer Tsachor  Online Weighted Paging with Distributions  at 2023-08-09 11:00:00
DESCRIPTION:We study the classic problem of online weighted paging with a probabilistic prediction model, in which we are given additional information about the input in the form of distributions overpage requests, known as distributional online paging (DOP).  Our main result is an efficient online algorithm that achieves a constant factor competitive ratio with respect to the best online algorithm (policy) for weighted DOP.Our starting point is a linear programming formulation for weighted DOP. Unfortunately, this formulation has a large integrality gap which depends on page weights. In our work we overcome the integrality gap by incorporating an additional rounding step based on dynamic programming, achieving the desired constant competitive factor algorithm for the problem.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture: 3906204304 and Taub 401
UID:eventx6a5a287eeb67710476
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20230810T150000
DTEND;TZID=Asia/Jerusalem:20230810T160000
DTSTAMP;TZID=Asia/Jerusalem:20230810T150000
SUMMARY: MSC  talk by Tal Swisa  Knowledge-Based Generalization of Event Chains  at 2023-08-10 15:00:00
DESCRIPTION:The script theory in psychology suggests that cognitive scripts, sequences of expected actions in commonly encountered situations, play a significant role in shaping our comprehension of the world. The concept of scripts was utilized in artificial intelligence in its early days, serving as a tool for representing procedural knowledge and enhancing story understanding. Script-based methods, like the Script Applier Mechanism (SAM), marked a significant advancement in AI, but their reliance on manually crafted rules posed scalability issues that led to the discontinuation of their use.In this seminar, we introduce a novel method that builds upon the concept of scripts to automatically create explicit schemes of event chains, which we refer to as Generalized Narratives (GNs). Our method uses knowledge graphs to transform sequences of events into GNs, creating a more abstract and generalized representation of events. This approach is completely explicit and transparent, resource-efficient, and does not require large training datasets or specialized hardware.In the seminar we will discuss the method in detail, including how we define and create Generalized Events, which are the building blocks of GNs, and how we use knowledge graphs to generalize event components. We will also discuss how our method can predict missing events in a given narrative.An evaluation of our method on a missing event prediction task reveals it as a strong competitor to a large language model baseline. This underscores that our method, while providing unique advantages like explicitness, transparency, and resource efficiency, distinct from deep learning, also holds its own in performance across many cases.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture: 98312219029 and Taub 601
UID:eventx6a5a287eeb69410477
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20230829T113000
DTEND;TZID=Asia/Jerusalem:20230829T123000
DTSTAMP;TZID=Asia/Jerusalem:20230829T113000
SUMMARY: MSC  talk by Tsachi Blau  Threat Model-Agnostic Adversarial Defense Using Diffusion Models  at 2023-08-29 11:30:00
DESCRIPTION:Deep Neural Networks (DNNs) are highly sensitive to imperceptible malicious perturbations, known as adversarial attacks. Following the discovery of this vulnerability in real-world imaging and vision applications, the associated safety concerns have attracted vast research attention, and many defense techniques have been developed. Most of these defense methods rely on adversarial training (AT) -- training the classification network on images perturbed according to a specific threat model, which defines the magnitude of the allowed modification. Although AT leads to promising results, training on a specific threat model fails to generalize to other types of perturbations. A different approach utilizes a preprocessing step to remove the adversarial perturbation from the attacked image. In this work, we follow the latter path and aim to develop a technique that leads to robust classifiers across various realizations of threat models. To this end, we harness the recent advances in stochastic generative modeling, and means to leverage these for sampling from conditional distributions. Our defense relies on an addition of Gaussian i.i.d noise to the attacked image, followed by a pretrained diffusion process -- an architecture that performs a stochastic iterative process over a denoising network, yielding a high perceptual quality denoised outcome. The obtained robustness with this stochastic preprocessing step is validated through extensive experiments on the CIFAR-10 dataset, showing that our method outperforms the leading defense methods under various threat models.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture: 96760244696 
UID:eventx6a5a287eeb6a610483
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20230905T113000
DTEND;TZID=Asia/Jerusalem:20230905T123000
DTSTAMP;TZID=Asia/Jerusalem:20230905T113000
SUMMARY: pixel-club  talk by Matar Tzur (Buchbinder)  Pixel Club:Cataract Retinal Image Quality Assessment using Deep Learning for Glaucoma Diagnosis  at 2023-09-05 11:30:00
DESCRIPTION:Glaucoma is a group of eye diseases that gradually leads to peripheral vision loss and blindness. It is affecting about 90 million people worldwide and usually painless. Glaucoma has no cure and it advances moderately if not treated on time. Therefore early and fast diagnosis is crucial for effective treatment. In this work I examine to what extent inferior retinal images affects glaucoma diagnosis. Then I develop a Retinal Image Quality Assessment (RIQA) system accordingly, to screen out irrelevant images for diagnosis. For this purpose cataract hazed images are chosen, since cataract is the main blindness cause in the western world and is highly common in older ages. Cataract simulated images are created using the Dark Channel Prior (DCP) technique. For the second step I perform dehazing algorithm on the cataract retinal images and test diagnosis results again, to assess the dehazing algorithm. I conclude at what cataract level glaucoma diagnosis is just unreliable, even after attempts to improve image quality by pre-processing.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture: 94712379821
UID:eventx6a5a287eeb6b910485
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20230906T113000
DTEND;TZID=Asia/Jerusalem:20230906T123000
DTSTAMP;TZID=Asia/Jerusalem:20230906T113000
SUMMARY: MSC  talk by  Sajy Khashab  Multitenant In-Network Acceleration with SwitchVM  at 2023-09-06 11:30:00
DESCRIPTION:In-Network Computing is a concept of acceleration of applications by offloading some computation to run on network devices. Recently enabled by the emergence of data-plane programmable PISA switches, in-network computing was shown to offer dramatic performance boosts in a variety of applications such as load balancers, coordination protocols, aggregation and more. However, existing switches lack the essential support for multitenancy, limiting the benefits only to data center operators.We propose a practical approach to implementing multitenancy on programmable network switches to make in-network acceleration accessible to cloud users. We introduce a Switch Virtual Machine (SwitchVM), that is deployed on the switches and offers an expressive instruction set and program state abstractions. Tenant programs, called data-plane filters (DPFs), are loaded on a per-packet basis and are executed on top of SwitchVM in a per-tenant sandbox with memory, network and state isolation policies controlled by network operators. The packets that trigger DPF execution include the code to execute or a reference to the DPFs deployed in the switch. DPFs are Turing-complete, may maintain state in the packet and in switch virtual memory, may form a dynamic chain, and may steer packets to desired destinations, all while enforcing the operator’s policies.We demonstrate that this idea is practical by prototyping SwitchVM in P4 on Intel Tofino switches. We describe a variety of use cases that SwitchVM supports, and implement three complex applications from prior works – Key-Value Store cache, Load-aware load balancer and Paxos accelerator. We also show that SwitchVM provides strong performance isolation, zero-overhead runtime programmability, may hold two orders of magnitude more in-switch programs than existing techniques, and may support up to thirty thousand concurrent tenants each with its private state.Joint work with Alon Rachelbach and Prof. Mark Silberstein. Paper will appear in NSDI '24.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture: 97991655601 and Taub 401 
UID:eventx6a5a287eeb6c910480
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20230907T090000
DTEND;TZID=Asia/Jerusalem:20230907T100000
DTSTAMP;TZID=Asia/Jerusalem:20230907T090000
SUMMARY: MSC  talk by Emil Barel  Correlation Clustering and Ratio Steiner Cuts in Hypergraphs  at 2023-09-07 09:00:00
DESCRIPTION:We introduce a new family of clustering problems, which we denote by Hyper Correlation Clustering, that takes into account higher-order structures.\nOur new family captures multiple classic graph cut problems, e.g., Min s-t Cut, Multiway Cut, and Multicut, in addition to disagreement minimization on general weighted graphs in Correlation Clustering, as well as other studied hypergraph clustering problems.In Hyper Correlation Clustering we are given a hypergraph H=(V,E) whose every hyperedge e is labeled by + or -, where + (-) indicates similarity (dissimilarity) of vertices in e.\nEach hyperedge is associated with an agreement function, that depends on its label, which quantifies the agreement of the hyperedge as a function of the number of times it is cut.The goal is to find a clustering that minimizes the total disagreement of hyperedges. While no reasonable approximation is possible for arbitrary agreement functions, we present a polylogarithmic approximation for general weighted hypergraphs and a natural class of agreement functions.Our algorithm is based on the combination of a combinatorial greedy approach and an approximation algorithm for a hypergraph cut problem that is closely related to both Sparsest Cut with general demands and Min Ratio Steiner Cut in which: (1) cost is over hyperedges as opposed to edges; and (2) demands are given over large subsets of vertices as opposed to pairs of vertices. We believe this hypergraph cut problem might be of independent interest.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture: 4356187325 
UID:eventx6a5a287eeb6dc10487
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20230910T100000
DTEND;TZID=Asia/Jerusalem:20230910T110000
DTSTAMP;TZID=Asia/Jerusalem:20230910T100000
SUMMARY: MSC  talk by Chiara Meiohas  Linear-Mark: Locality vs. Accuracy in Mark-Sweep Garbage Collection  at 2023-09-10 10:00:00
DESCRIPTION:Tracing garbage collectors are widely deployed in modern programming languages. But tracing an arbitrary heap shape incurs poor locality and may hinder scalability. In this paper, we explore an avenue for mitigating these inefficiencies at the expense of conservative, less accurate identification of live objects. We do this by proposing and studying an alternative to the Mark-Sweep tracing algorithm, called Linear-Mark. It turns out that although Linear-Mark improves locality and scalability, the accuracy of Mark-Sweep outweighs the achieved enhancements. We present the Linear-Mark garbage-collecting algorithm and provide an evaluation that highlights the trade-offs between the Linear-Mark and the Mark-Sweep approaches. Our hope is that this research will inspire further algorithmic improvements, ultimately leading to better garbage collection algorithms.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture: 2945058890 and Taub 401
UID:eventx6a5a287eeb6eb10484
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20230912T113000
DTEND;TZID=Asia/Jerusalem:20230912T123000
DTSTAMP;TZID=Asia/Jerusalem:20230912T113000
SUMMARY: PHD  talk by Eyal Rozenberg  Reduced-Supervision and Ab-initio Machine Learning Approaches with Applications to Quantum Physics and Medicine  at 2023-09-12 11:30:00
DESCRIPTION:This research is centered around the application of machine learning approaches to domains in which training labels are prohibitively expensive to obtain or training data in general are impossible to procure.Initially, the study focused on investigating weak supervision in medical applications, in which data labelling is extremely expensive; specifically showcasing how a restricted number of high-quality labels can significantly improve algorithm performance in diagnosing chest diseases from X-ray images. Subsequently, the research shifted its emphasis towards tackling intricate problems in physics, such as quantum chemistry and quantum optics, which inherently encounter difficulties in database creation.The challenges in physics pertain to integrating physical constraints into machine learning models to confine the model's search space, thereby mitigating the need for extensive experiments or simulations. An initial study addresses a problem in quantum chemistry that revolves around determining the ground state energy of molecules. We consider an ab-initio quantum chemistry method, which involves incorporating the model with all pertinent physical constraints of the problem and enabling it to converge towards the wave function that represents the electron distribution around the molecule.A second research endeavor in physics centers on employing advanced computational learning tools to address inverse design problems in quantum optics to obtain the desired high dimensional bi-photon entanglement. The objective is to qualitatively represent the physical model in a differentiable manner, particularly focusing on the non-linear interaction of light and matter. Importantly, this is achieved without relying on any empirical or theoretical data that contains the design of the optical system and the resulting entangled photons. We further validate our approach against experimental results.\n
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture: 99665032956 
UID:eventx6a5a287eeb6fb10482
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20230913T113000
DTEND;TZID=Asia/Jerusalem:20230913T123000
DTSTAMP;TZID=Asia/Jerusalem:20230913T113000
SUMMARY: ceClub  talk by Alon Rashelbach  ceClub: Trading Memory Accesses for Computations in Packet Processing and Beyond  at 2023-09-13 11:30:00
DESCRIPTION:Range matching plays a crucial role in computer systems, including networking, security, and storage. It serves the purpose of locating a range that encompasses a given input number from a vast collection of ranges. Address translators in operating systems and longest-prefix matching in networks heavily rely on range matching. However, existing range matching algorithms are limited in scalability and performance due to their reliance on pointer-chasing techniques.We introduce a novel data structure called the Range Query Recursive Model Index (RQRMI) to address the challenges associated with range matching. By leveraging shallow neural networks, RQRMI enables the learning of range distributions, transforming expensive lookup operations into efficient neural network inference. By employing the RQRMI model, we achieve an impressive range compression ratio of up to 90X. This compression capability allows for direct lookup operations while fitting within the CPU core cache. Importantly, the RQRMI training algorithm guarantees a strict upper bound on lookup latency, ensures the correctness of results, and exhibits fast convergence rates.\nWe have developed NuevoMatch, an algorithm for multi-field packet classification that leverages RQRMI models (SIGCOMM’20), and successfully integrated it into the critical path of Open vSwitch, a broadly used virtual switch (NSDI’22). Through the utilization of RQRMI, we have achieved remarkable scalability, enabling Open vSwitch to handle 500 times more routing rules while experiencing a throughput speedup of up to 160 times. Furthermore, our work demonstrates the versatility of RQRMI beyond software. Specifically, we observed up to 20X reduction in the memory footprint required for DNA hardware accelerators (BCB’23) and developed hardware for enhancing the scalability of network packet processors (MICRO’23).Joint work with Igor DePaula, Ori Rottenstreich, and Mark SilbersteinAlon is a PhD student supervised by Prof. Mark Silberstein and Prof. Ori Rottenstreich.\n
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Room 861, EE Meyer Building & Zoom Lecture: 97150849786
UID:eventx6a5a287eeb70b10488
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20230926T140000
DTEND;TZID=Asia/Jerusalem:20230926T150000
DTSTAMP;TZID=Asia/Jerusalem:20230926T140000
SUMMARY: MSC  talk by Hen Kas-Sharir  Efficient Concurrent Size  at 2023-09-26 14:00:00
DESCRIPTION:Determining the size of a concurrent data structure correctly and efficiently in the presence of concurrent modifications has turned out to be a surprisingly difficult task, one that has been absent from both research and practical applications until recently. In this work, we study three methodologies for concurrently computing a linearizable size, with the aim of improving performance. In our first approach, we employ the handshake methodology used by on-the-fly garbage collectors. In the second we use efficient locking mechanisms, while in the final methodology we use an an optimistic approach. Our study reveals that there is no one-size-fits-all solution for all scenarios. Consequently, we offer recommendations tailored to possible scenarios, guiding the selection of the most suitable method.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 601
UID:eventx6a5a287eeb71d10490
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20230928T090000
DTEND;TZID=Asia/Jerusalem:20230928T100000
DTSTAMP;TZID=Asia/Jerusalem:20230928T090000
SUMMARY: MSC  talk by Yarden Adir  Approximation of Hierarchical Clustering  at 2023-09-28 09:00:00
DESCRIPTION:The hierarchical clustering problem deals with the construction of an M-layer hierarchical partition of a given graph. Every pair of vertices in the graph is associated with a layer. The objective is to construct a hierarchical partition that separates vertices as close as possible to their associated layer. It is proven that any approximation algorithm for this problem, induces an approximation algorithm of the same factor, for the problem of fitting tree metrics to general data, so as to minimize the additive error. The need to fit tree metrics to general data arises in several disciplines, such as in numerical taxonomy, in which fitting a tree metric to the data, could assist in learning the evolution tree. We consider two special cases of the hierarchical clustering problem. The first, is the sequential multicut problem, which is the hierarchical version of the well-known multicut problem. The second, mainly assumes the given graph is a tree. We present an O(log n)-approximation algorithm for the first case, and a bi-criteria, O(1)-approximation algorithm for the second.\n
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture: 9235239856 
UID:eventx6a5a287eeb72c10491
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20230928T140000
DTEND;TZID=Asia/Jerusalem:20230928T150000
DTSTAMP;TZID=Asia/Jerusalem:20230928T140000
SUMMARY: MSC  talk by Alon Mor  Accelerating the Global Aggregation of Local  at 2023-09-28 14:00:00
DESCRIPTION:Local explanation methods highlight the input tokens that have a considerable impact on the outcome of classifying the document at hand. For example, the Anchor algorithm applies a statistical analysis of the sensitivity of the classifier to changes in the token. Aggregating local explanations over a dataset provides a global explanation of the model. Such aggregation aims to detect words with the most impact, giving valuable insights about the model, like what it has learned in training and which adversarial examples expose its weaknesses. However, standard aggregation methods bear a high computational cost: a na"ive implementation applies a costly algorithm to each token of each document, and hence, it is infeasible for a simple user running in the scope of a short analysis session.\nWe devise techniques for accelerating the global aggregation of the Anchor algorithm. Specifically, our goal is to compute a set of top-k words with the highest global impact according to different aggregation functions. Some of our techniques are lossless and some are lossy.\nWe show that for a very mild loss of quality, we are able to accelerate the computation by up to 30$\times$, reducing the computation from hours to minutes. We also devise and study a probabilistic model that accounts for noise in the Anchor algorithm and diminishes the bias toward words that are frequent yet low in impact.\n
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture: 4721546630 
UID:eventx6a5a287eeb73c10489
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20231011T113000
DTEND;TZID=Asia/Jerusalem:20231011T123000
DTSTAMP;TZID=Asia/Jerusalem:20231011T113000
SUMMARY: PHD  talk by Avi Mizrahi  Coding Schemes for Blockchain Networks  at 2023-10-11 11:30:00
DESCRIPTION:We study the design of coding schemes for blockchain networks, focusing on state organization and memory-efficient data structures for communication protocols. We first propose traffic-aware sharding, a technique that arranges data into distinct groups, to decrease overhead from cross-shard transactions while providing memory-efficient mappings of data into shards. Then, we study the use of Merkle trees in transaction proof verification, discussing a traffic-aware approach to organize data in such trees for cost-effective communication. We also study the Invertible Bloom Lookup Table (IBLT), a probabilistic data structure used in set reconciliation protocols such as the synchronization of blockchain transaction mempools. Beyond the basic Bloom filter that can answer membership queries, the IBLT has a listing operation that traditionally succeeds in a probabilistic manner. We suggest an IBLT with listing guarantees, in which listing always succeeds when the size of the represented set is smaller than a given threshold. The constructions are based on various coding techniques such as stopping sets of error-correcting codes, Steiner systems, as well as new methodologies we develop. We provide an in-depth analysis of the parameter space of each of the constructions.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture: 94964184897 
UID:eventx6a5a287eeb74e10492
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20231017T113000
DTEND;TZID=Asia/Jerusalem:20231017T123000
DTSTAMP;TZID=Asia/Jerusalem:20231017T113000
SUMMARY: pixel-club  talk by Omer Yair  Pixel Club: Understanding, Improving, And Extending The Contrastive Divergence Method For Training Energy-Based Models  at 2023-10-17 11:30:00
DESCRIPTION:Recent years have witnessed remarkable advancements in generative models within the realm of computer vision. However, while great progress has been made in implicit generative techniques (e.g. GANs and Diffusion Models), methods that explicitly model the data distribution have been significantly lagging behind. This seminar will present our research on such methods, which are collectively known as Energy-Based Models (EBMs). I will start by revisiting the classical Contrastive Divergence algorithm for training EBMs (Hinton, 2002). The original derivation of this algorithm relied on an unjustified approximation. Here, I will show that this method can be derived in an alternative way, which relies on no approximations, and sheds new light on how and why the CD algorithm works. Based on insights from our analysis, I will then present an improved CD method that substantially narrows the performance gap to the current state-of-the-art techniques. Finally, I will demonstrate how our method can be harnessed for visualizing uncertainties in inverse problems.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Room 1061, EE Meyer Building & Zoom Lecture: 92473792641
UID:eventx6a5a287eeb76010495
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20231022T133000
DTEND;TZID=Asia/Jerusalem:20231022T143000
DTSTAMP;TZID=Asia/Jerusalem:20231022T133000
SUMMARY: MSC  talk by Noam Rotstein  Multimodal Image Mappings: From Geometric Alignment to Captions  at 2023-10-22 13:30:00
DESCRIPTION:At the heart of our research lies a simple yet profound question: how can we bridge the gap between different visual, geometric, and language modalities? Firstly, we address the task of aligning colored point clouds embedded in 3D, obtained by a colored depth scanner, with color images provided by conventional cameras. These two data forms are inherently different, in both structural and chromatic properties. We use a tailored optimization procedure to align the point cloud and camera image by reducing photometric discrepancies between them. This ensures precise modality correspondence, resulting in RGBD data that enhances geometric reconstruction. In our second effort, we focus on the relation between images and text. While vision-language pre-training (VLP) has significantly advanced image captioning, models tend to provide generic descriptions, omitting salient details. This issue stems from training datasets that, while capturing broad image content, frequently overlook specifics. To address this, we enrich captions with insights from "frozen" vision experts like object detectors and attribute extractors. Our method, FuseCap, fuses this data with original captions via a large language model (LLM), producing a vast dataset of detailed captions. Models trained on this data offer enhanced performance and richer descriptions. Beyond intermodality, a core principle across both studies is a data-centric AI strategy. By focusing on data quality rather than exhaustive model refinement, we significantly enhance the outcomes of data-hungry models.\n
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture: 93548839223 
UID:eventx6a5a287eeb77110497
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20231115T110000
DTEND;TZID=Asia/Jerusalem:20231115T120000
DTSTAMP;TZID=Asia/Jerusalem:20231115T110000
SUMMARY: CSpecial Event  talk by Prof. Eitan Yaakobi   Information Storage Systems: Past, Present And Back To The Future  at 2023-11-15 11:00:00
DESCRIPTION:In the current period, we invite you to "eye-level" scientific intermission lectures especially for you. The lectures will be delivered online by our faculty members.\nThe first lecture: "Information storage systems: past, present and back to the future"\nProf. Eitan Yaakobi \nWednesday 11/15 at 11:00\n\nwaiting for you!!
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture: 98560245665 
UID:eventx6a5a287eeb78510501
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20231115T130000
DTEND;TZID=Asia/Jerusalem:20231115T140000
DTSTAMP;TZID=Asia/Jerusalem:20231115T130000
SUMMARY: MSC  talk by Omer Belhasin  TransBoost: Improving The Best ImageNet Performance using Deep Transductive Learning  at 2023-11-15 13:00:00
DESCRIPTION:This lecture is about our paper that was published in NeurIPS 2022. This paper deals with deep transductive learning, and proposes TransBoost as a procedure for fine-tuning any deep neural model to improve its performance on any (unlabeled) test set provided at training time. TransBoost is inspired by a large margin principle and is efficient and simple to use. Our method significantly improves the ImageNet classification performance on a wide range of architectures, such as ResNets, MobileNetV3-L, EfficientNetB0, ViT-S, and ConvNext-T, leading to state-of-the-art transductive performance. Additionally we show that TransBoost is effective on a wide variety of image classification datasets.\n
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture: 93361356250 
UID:eventx6a5a287eeb79610500
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20231116T113000
DTEND;TZID=Asia/Jerusalem:20231116T123000
DTSTAMP;TZID=Asia/Jerusalem:20231116T113000
SUMMARY: MSC  talk by Shachar Katz  Visualizing and Interpreting the Semantic Information Flow of Transformers  at 2023-11-16 11:30:00
DESCRIPTION:Recent advances in interpretability research suggest we can project weights and hidden states of transformer-based language models (LMs) to their vocabulary, a transformation that makes them more human interpretable. In this paper, we investigate LM attention heads and memory values, the vectors the models dynamically create and recall while processing a given input. By analyzing the tokens they represent through this projection, we identify patterns in the information flow inside the attention mechanism. Based on our discoveries, we create a tool to visualize a forward pass of Generative Pre-trained Transformers (GPTs) as an interactive flow graph, with nodes representing neurons or hidden states and edges representing the interactions between them. Our visualization simplifies huge amounts of data into easy-to read plots that can reflect the models’ internal processing, uncovering the contribution of each component to the models’ final prediction. Our visualization also unveils new insights about the role of layer norms as semantic filters that influence the models’ output, and about neurons that are always activated during forward passes and act as regularization vectors. 
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 601
UID:eventx6a5a287eeb7a810498
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20231121T103000
DTEND;TZID=Asia/Jerusalem:20231121T113000
DTSTAMP;TZID=Asia/Jerusalem:20231121T103000
SUMMARY: MSC  talk by Noam Ben-Moshe  Machine learning for atrial fibrillation analysis from the raw ECG waveform  at 2023-11-21 10:30:00
DESCRIPTION:Atrial fibrillation (AF) is the most prevalent form of heart arrhythmia and is associated with a fivefold increase in stroke incidence. In the context of AF detection, some patients experience sporadic AF events. This makes the Holter electrocardiogram (ECG) examination, which captures longer-term heart activity, essential to capture these irregular events. Automatic detection of AF in Holter recordings has the potential to reduce clinician workload. On the ECG, AF is characterized by an irregular rhythm and by the presence of fibrillatory waves (f-wave). While the standard Holter examination typically includes three leads, single-lead ECGs have become increasingly more common thanks to the development of patches and smartwatches for remote health monitoring and screening. This research makes two scientific contributions towards creating AI driven systems to support the detection and analysis of AF in single lead ECG. First, we propose a new method for ranking f-wave extraction methods. Second, we develop a robust, i.e., highly performing and generalizable model, for AF events detection and benchmark this new model to state-of-the-art.The new method for ranking f-wave extraction algorithms is based on the hypothesis that better-performing AF classification using features computed from the extracted f-waves implies better-performing extraction. Three independent Holter datasets and four f-wave extraction algorithms were used for this experiment. The results showed that the PCA-based f-wave extraction approach was superior on all datasets and for all leads. A significant advantage of our evaluation method lies in its ability to leverage real datasets without the need for ground truth f-waves.The elaboration of robust algorithms for AF events detection from single lead ECG present several challenges. These include distribution shifts across lead, hardware, ethnicity and the inherent presence of noise within the ECG signal. A new deep learning model, called RawECGNet, was developed for the task of AF detection. In order to enhance the generalization capabilities of the model, it was trained on single lead ECG input from different lead position. In addition, the domain shift uncertainty (DSU) layer was included to introduce a degree of uncertainty into the encoding process. RawECGNet demonstrates exceptional generalization both in the source domain and two target domains and outperformed a state-of-the-art deep learning model taking as input the beat-to-beat interval time series. RawECGNet harnesses the complete potential of ECG waveform morphology for enhanced diagnostic accuracy.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture: 94193068004 and Taub 401
UID:eventx6a5a287eeb7bb10499
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20231122T113000
DTEND;TZID=Asia/Jerusalem:20231122T123000
DTSTAMP;TZID=Asia/Jerusalem:20231122T113000
SUMMARY: CSpecial Event  Advocate For The Release Of The Abductees  at 2023-11-22 11:30:00
DESCRIPTION:On Wednesday, November 22, 2023, at 11:30 AM, we will gather to advocate for the release of the abductees. This event will be held at the Central Library Square.Due to the Home Front Command's regulations on public gatherings, registration is required to participate.Please register at the following link.Together, we stand strong!
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:the Central Library Square
UID:eventx6a5a287eeb7d410504
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20231123T163000
DTEND;TZID=Asia/Jerusalem:20231123T173000
DTSTAMP;TZID=Asia/Jerusalem:20231123T163000
SUMMARY: PHD  talk by Boris Pismenny  Autonomous application offloads using network controllers  at 2023-11-23 16:30:00
DESCRIPTION:The Internet services which we enjoy in our day-to-day lives---search, social networking, online maps, video sharing, online shopping—run on Data Centers (DCs). DCs are warehouse scale computers that consist of tens of thousands of machines which are interconnected via fast networks. Building and maintaining DCs is tremendously expensive, for example, Amazon’s DC in Tel-Aviv spans over 100,000 square feet and they estimate that building each DC costs approximately 2.37 billion USD where only 280 million USD are designated for the land and buildings and the rest is for computing infrastructure. Technology companies seeking to maximize their return on investment (ROI) must efficiently utilize their DCs which is particularly challenging because rapid computer technology changes shift the bottleneck component in DCs. In our work, we observe that in recent years the growth in host Network Interface Controller (NIC) bandwidth has outpaced the growth in other DC host system resources such as memory bandwidth and CPU processing capacity. At the same time, networking is the cheapest component in servers whereas CPU and memory are the most expensive, therefore finding new ways to improve the utilization of NICs will improve overall DC utilization and ROI.Our first paper “Autonomous NIC Offloads” tackles the problem of offloading CPU-intensive application layer logic (e.g., encryption) onto NICs. We observe that the ideal position to perform data-intensive computations is on the NIC as network data flows through it in any case. But previous approaches to offload data-intensive application layer (layer-5) computations to NICs depend on offloading the underlying layer≤4 protocols (TCP, IP, routing, firewall, etc.), which undesirably encumbers innovation imposing undesirable security and maintenance burdens. In contrast, autonomous NIC offloads accelerate data-intensive application logic without having to migrate the entire layer≤4 network stack to the NIC. The key challenges autonomous offloads address is coping with out-of-sequence TCP packets. On transmit, to process out-of-sequence packet P, we leverage the software TCP retransmission buffer and the application itself to provide the NIC with the data needed to process P. On receive, out-of-sequence packets bypass the offloading logic when the NIC’s state is insufficient to perform the offload, and we use a software-hardware handshake to recover the state necessary to offload subsequent packets. We implement autonomous offloads for two protocols and computations: HTTPS encryption and authentication and NVMe-TCP zero-copy and data digest. We also describe the properties of protocols and computations that are autonomously offloadable, we find that most are offloadable but not all. Our evaluation shows autonomous offloads increase throughput by up to 3.3x, reduce CPU utilization by up to 60% and reduce latency by up to 30%. Software support for autonomous offloads is available in open-source libraries, such as the Linux Kernel and OpenSSL, and recent NVIDIA NICs support autonomous offloads in hardware. Our second paper “The Benefits of General-Purpose On-NIC Memory”exposes the newly available memory on NICs (Nicmem) directly to applications. We identify a class of applications that benefit from Nicmem, which we call “data movers”. Data mover applications process incoming packets based on metadata without accessing incoming packet data. We use two data mover applications to demonstrate the benefits of Nicmem: key-value stores (KVS) and network functions (NFs). Popular NFs such as network address translation frequently operate on headers—rather than data—of incoming packets. For NFs, we introduce a packet processing architecture that splits between packet headers and data, keeping the data on Nicmem when possible and thus reducing memory and PCIe bandwidth. Our approach consequently shortens latency by up to 23% and increases throughput by up to 19%. Similarly, because KVS workloads are highly skewed, we introduce a cache of hot values that resides on Nicmem which is closer to the wire. This design shortens skewed KVS workload latency by up to 43% and increases throughput by up to 80%.Our third paper “ShRing: Networking with Shared Receive Rings” observes that today’s NIC interface for receiving packets requires sufficient per-core packet buffers to absorb packet bursts, but the combined size of all packet buffers—which are typically not shared between cores—can exceed the size of the last level cache (LLC). As a result, packet processing slows down, degrading throughput and latency, because NIC and CPU memory accesses are frequently served from main memory rather than LLC. To alleviate this problem, we propose a new NIC interface for receiving packets called “shRing” which shares packet buffers between cores when memory bandwidth is high. Inter-core sharing adds synchronization overhead which is offset by the smaller memory footprint. Our experiments show that shRing increases NF throughput by up to 1.27x and reduces NF latency by up to 38x. The large latency improvement occurs when shRing reduces packet processing time below packet inter-arrival time thereby preventing CPU overload and queue buildup.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture: 2354687029 
UID:eventx6a5a287eeb7e810494
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20231128T170000
DTEND;TZID=Asia/Jerusalem:20231128T200000
DTSTAMP;TZID=Asia/Jerusalem:20231128T170000
SUMMARY: CSpecial Event  First Aid Workshop  at 2023-11-28 17:00:00
DESCRIPTION:You are invited to register for a first aid workshop that will be held at the faculty at the initiative of the student council. \n\nTuesday 11/28 at 17:00, at the multi-purpose center in Taub.\nPre-registration is required at the link (the number of participants is limited)\n\nThe workshop will deal with providing first aid in an emergency and will provide tools and ways of dealing with injuries and common medical emergencies.\nThe contents are suitable for anyone who wants to acquire tools and knowledge and help save lives and do not require prior knowledge and/or medical training.\n\nThe workshop is free of charge and will be conducted by MDA volunteer instructors.\nDuration of the workshop - three hours.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:The Multi-Purpose Center, Floor 0
UID:eventx6a5a287eeba3310506
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20231129T103000
DTEND;TZID=Asia/Jerusalem:20231129T113000
DTSTAMP;TZID=Asia/Jerusalem:20231129T103000
SUMMARY: MSC  talk by Mohammad Masarwy  Value‬‬ ‫‪of‬‬ ‫‪Assistance‬‬ ‫‪for‬‬ ‫‪Grasping‬‬  at 2023-11-29 10:30:00
DESCRIPTION:In many realistic settings, a robot is tasked with grasping an object without knowing the object's exact pose. Instead, the robot relies on a probabilistic estimation of the object pose to decide how to attempt to grasp the object. We offer a novel measure, called Value of Assistance, or VOA, for assessing the expected effect a specific observation will have on the robot's ability to successfully grasp the object. VOA supports the decision of where and when it would be most beneficial to perform a sensing action, such as taking a picture of the scenario, to the grasping task at hand.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture: 5713866909 and Taub 601
UID:eventx6a5a287eeba5210505
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20231204T133000
DTEND;TZID=Asia/Jerusalem:20231204T143000
DTSTAMP;TZID=Asia/Jerusalem:20231204T133000
SUMMARY: PHD  talk by  Victor Kolobov  Homomorphic Secret Sharing and Information Theoretic Cryptography  at 2023-12-04 13:30:00
DESCRIPTION:Our research focuses on new techniques for homomorphic secret sharing (HSS) which is a promising new cryptographic tool for privacy-preserving computations. HSS can be seen as a relaxation of fully homomorphic encryption (FHE), the latter being an encryption with the capability to perform calculations on encrypted data without decrypting first.FHE is a well-studied topic in cryptography that has recently attracted a lot of research both in academia and in the industry. However, the efficiency of state-of-the-art solutions leaves much to be desired. HSS relaxes the notion of FHE by allowing the data to be secret-shared among two or more non-colluding servers, as opposed to being encrypted on a single server.HSS has several advantages over FHE. On the theory side, non-trivial HSS schemes exist based on symmetric (“private key”) cryptography or even with information-theoretic (“perfect”) security. On the practical side, these low-end HSS solutions are orders of magnitude faster than the single-server FHE solutions and have better concrete communication costs. Our research explores new approaches to constructing efficient HSS schemes by improving the generality, communication complexity, and computational cost of existing techniques. As a special case, this has relevance to multi-server private information retrieval (PIR), a form of HSS for table lookup which is a powerful building block for privacy-preserving database searches.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture: 93797137886 and Taub 301
UID:eventx6a5a287eeba6810502
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20231205T113000
DTEND;TZID=Asia/Jerusalem:20231205T123000
DTSTAMP;TZID=Asia/Jerusalem:20231205T113000
SUMMARY: pixel-club  talk by Omer Yair  Understanding, Improving, And Extending The Contrastive Divergence Method For
Training Energy-Based Models  at 2023-12-05 11:30:00
DESCRIPTION:Recent years have witnessed remarkable advancements in generative models within the realm of computer vision. However, while great progress has been made in implicitgenerative techniques (e.g. GANs and Diffusion Models), methods that explicitly model the data distribution have been significantly lagging behind. This seminar will present our research on such methods, which are collectively known as Energy-Based Models (EBMs). I will start by revisiting the classical Contrastive Divergence algorithm for training EBMs (Hinton, 2002). The original derivation of this algorithm relied on an unjustified approximation. Here, I will show that this method can be derived in an alternative way, which relies on no approximations, and sheds new light on how and why the CD algorithm works. Based on insights from our analysis, I will then present an improved CD method that substantially narrows the performance gap to the current state-of-the-art techniques. Finally, I will demonstrate how our method can be harnessed for visualizing uncertainties in inverse problems.\n\nPh.D. Under the supervision of Prof. Tomer Michaeli.\n
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Room 1061, EE Meyer Building 
UID:eventx6a5a287eeba7e10507
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20231214T110000
DTEND;TZID=Asia/Jerusalem:20231214T120000
DTSTAMP;TZID=Asia/Jerusalem:20231214T110000
SUMMARY: MSC  talk by Rani Abboud  Practical Heavy-Hitter Detection Algorithms for Programmable Switches  at 2023-12-14 11:00:00
DESCRIPTION:Programmable switches enable offloading various network functions, such as anomaly detection and traffic engineering, to the same switches that perform packet routing. A basic component in many such applications is detecting heavy hitters (largest flows).Realizing such data plane algorithms requires taking into consideration all types of limited hardware resources of the switch, including the recirculation bandwidth, number of stages, and memory. This motivates solutions that avoid recirculation when possible and fit into a minimal number of stages, even at the cost of slightly higher memory usage, since memory is not a tight resource on modern programmable switches.We introduce a novel sketch for heavy hitter detection, CMSIS, that supports both online detection and offline retrieval of heavy hitters and achieves high accuracy while incurring low resource consumption. In particular, CMSIS requires no recirculation and consumes 25% less pipeline stages than state-of-the-art alternatives that do not perform recirculation, while its memory consumption is competitive with prior works.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 401
UID:eventx6a5a287eeba9210512
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20231214T123000
DTEND;TZID=Asia/Jerusalem:20231214T133000
DTSTAMP;TZID=Asia/Jerusalem:20231214T123000
SUMMARY: MSC  talk by Gal Avineri  Offline Meta-RL: Applicable Ambiguity Alleviation  at 2023-12-14 12:30:00
DESCRIPTION:In meta reinforcement learning (meta-RL) an agent seeks an optimal policy when facing a new unseen task that is sampled from a known task distribution. Such a policy leads an effective trade-off between information gathering and reward accumulation. The offline variant of meta-RL (OMRL) presents a challenge to learn such a policy, as previous work established an identifiability problem in OMRL termed MDP ambiguity. This problem relates to the difficulty of learning a neural network that can infer the task at hand at test time. We propose a new method to utilize prior knowledge of the task distribution to mitigate the identifiability problem in OMRL. Additionally, we propose a novel method to evaluate an inference model \textit{offline}, which is more efficient and accurate than the online alternative of policy optimization. Finally, we show that the offline version of the popular VariBAD algorithm can learn a suboptimal representation for task inference, and propose a simple modification that uses contrastive predictive coding to improve its performance. We compare our methods to Offline VariBAD on two ambiguity-prone tasks and demonstrate results that are on par or better than policy replay - a state of the art method for solving MDP ambiguity - while requiring weaker assumptions.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture: 94960036903 and Taub 601
UID:eventx6a5a287eebaa610510
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20231226T110000
DTEND;TZID=Asia/Jerusalem:20231226T120000
DTSTAMP;TZID=Asia/Jerusalem:20231226T110000
SUMMARY: PHD  talk by Hadas Biran  Statistical Methods for Analyzing RNA Sequencing Data: Structures, MicroRNA Activity and Point Mutations  at 2023-12-26 11:00:00
DESCRIPTION:Over the past two decades, advancements in gene expression laboratory methods have brought about a level of maturity that allows for the routine examination of gene expression at both the single-cell level and spatially across tissues. However, existing data analysis methods in single-cell sequencing predominantly concentrate on identifying cell clusters or delineating the principal progression line within the data. Spatial transcriptomics analysis primarily focuses on clustering and identifying spatially variable genes. While these methods effectively capture the primary features of the data, they may overlook more subtle processes, potentially involving specific subsets of samples. Also, widely employed techniques do not detect the regulatory activity of microRNAs, which play a crucial role in governing the expression of mRNAs and lncRNAs.In this talk, I will introduce SPIRAL, an algorithm grounded in Gaussian statistics that is adept at identifying all statistically significant biological processes in single-cell, bulk, and spatial transcriptomics data. I will also present miTEA-HiRes, a method designed to facilitate the evaluation of microRNA activity at a high resolution. Lastly, I will speak about our work in detecting somatic point mutations in bulk RNA-seq samples.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture: 3182949557 and Taub 601
UID:eventx6a5a287eebaba10509
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20240101T103000
DTEND;TZID=Asia/Jerusalem:20240101T113000
DTSTAMP;TZID=Asia/Jerusalem:20240101T103000
SUMMARY: colloq  talk by Yuval Dagan  CS Lecture: Learning From Dependent Data And Its Modeling Through The Ising Model  at 2024-01-01 10:30:00
DESCRIPTION:I will present a theoretical framework for analyzing learning algorithms which rely on dependent, rather than independent, observations. While a common assumption is that the learning algorithm receives independent datapoints, such as unrelated images or texts, this assumption often does not hold. An example is data on opinions across a social network, where opinions of related people are often correlated, for example as a consequence of their interactions. I will present a line of work that models the dependence between such related datapoints using a probabilistic framework in which the observed datapoints are assumed to be sampled from some joint distribution, rather than sampled i.i.d. The joint distribution is modeled via the Ising model, which originated in the theory of Spin Glasses in statistical physics and was used in various research areas. We frame the problem of learning from dependent data as the problem of learning parameters of the Ising model, given a training set that consists of only a single sample from the joint distribution over all datapoints. We then propose using the Pseudo-MLE algorithm, and provide a corresponding analysis, improving upon the prior literature which necessitated multiple samples from this joint distribution. Our proof benefits from sparsifying a model's interaction network, conditioning on subsets of variables that make the dependencies in the resulting conditional distribution sufficiently weak. We use this sparsification technique to prove generic concentration and anti-concentration results for the Ising model, which have found applications beyond the scope of our work.Based on joint work with Constantinos Daskalakis, Anthimos Vardis Kandiros, Nishanth Dikkala, Siddhartha Jayanti, Surbhi Goel and Davin Choo.Yuval Dagan is a postdoctoral researcher at the Simons Institute for the Theory of Computing at UC Berkeley and at the Foundations of Data Science Institute (FODSI). He received his PhD from the Electrical Engineering and Computer Science Department at MIT, advised by Professor Constantinos Daskalakis (2018-2023). He received his Bachelor’s and Master’s degrees from the Technion, where he was advised by Professor Yuval Filmus (2011-2017). During his PhD, he received the Meta Research Fellowship in Machine Learning (2021-2022). Further, he was a visitor of the Simons Foundation at the Causality program (2022) and a research intern at Google Mountain View, hosted by Vitaly Feldman (2019). Prior to his PhD, he was a research assistant of Professor Ohad Shamir at Weizmann Institute (2018).
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Auditorium 012, floor 0
UID:eventx6a5a287eebad110515
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20240104T123000
DTEND;TZID=Asia/Jerusalem:20240104T133000
DTSTAMP;TZID=Asia/Jerusalem:20240104T123000
SUMMARY: PHD  talk by Vadim Gliner  12-Lead ECG Classification Using Deep Learning Methods  at 2024-01-04 12:30:00
DESCRIPTION:12-lead electrocardiogram (ECG) recordings can be collected in any clinic and the interpretation is performed by a clinician. Modern machine learning tools may make them automatable. However, a large fraction of 12-lead ECG data is still available in printed paper or image only and comes in various formats. To digitize the data, smartphone cameras can be used. Nevertheless, this approach may introduce various artifacts and occlusions into the obtained images.Here, I will present 5 papers (3 published) that describe our journey toward a clinical trial. In our first paper, we designed an automated algorithm to classify short ECG signal strips into 4 categories: normal rhythm, atrial fibrillation, noisy segment, or other rhythm disturbances. We used a feature-based classification to classify the short ECG recordings. Our algorithm obtained a total score (F1) of 0.80 on the hidden dataset. Our algorithm was able to classify AF vs. non-AF and normal vs. abnormal (arrhythmia or noise) records. In our second paper, we introduce a two-way approach to an automated cardiac disease identification system using standard digital or image 12-lead ECG recordings. Two different network architectures, one trained using digital signals (CNN-dig) and one trained using images (CNN-ima), were generated. CNN-dig accurately (92.9-100%) identified every possible combination of the eight most-common cardiac conditions.  Both CNN-dig and CNN-ima accurately (98%) detected AF from standard 12-lead ECG digital signals and images, respectively. In our third paper, we overcome the challenges of automating 12-lead ECG analysis using mobile-captured images and a deep neural network that is trained using a domain adversarial approach. The net achieved an average 0.91 receiver operating characteristic curve on tested images captured by a mobile device. Assessment on image from unseen 12-lead ECG formats that the network was not trained on achieved high accuracy. The network accuracy can be improved by including a small number of unlabeled samples from unknown formats in the training data. Finally, our models achieve high accuracy using signals as input rather than images.In our fourth paper, we aim to introduce a high-resolution ECG interpretation tool designed for real clinical images captured by mobile devices. Our approach capitalizes on the sensitivity of the Jacobian matrix for input images. We showcase interpretability for both morphological and arrhythmogenic cardiac conditions in images captured in clinical environments. The interpretability tool accentuates key signal features with high resolution, aligning with known clinical signs.In our fifth paper, we present BeatBox AI—an automated diagnostic platform for 12 lead ECGs. This innovative system is crafted as a continually self-optimizing solution, offering scalability and adaptability to diverse populations and various 12-lead ECG layouts. BeatBox AI's diagnostic capabilities were assessed using 4,060 ECGs gathered during a 9-month prospective clinical study. This study involved collaboration with 14 cardiologists from five distinct hospitals. In the comparative analysis between the automatic interpretation and the cardiologists' assessments, the system demonstrated improvement across all operational dimensions. At the concluding evaluation stage, the MCC significantly improved to 0.56, and the system successfully identified a total of 54 distinct cardiac conditions.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Faculty of Biomedical Engineering, Silver Building, Room 201
UID:eventx6a5a287eebaeb10514
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20240116T110000
DTEND;TZID=Asia/Jerusalem:20240116T120000
DTSTAMP;TZID=Asia/Jerusalem:20240116T110000
SUMMARY: colloq  talk by Joachim Neu   Internet-Scale Consensus in the Blockchain Era  at 2024-01-16 11:00:00
DESCRIPTION:Blockchains have ignited interest in Internet-scale consensus as a vital building block for decentralized applications and services that promise egalitarian access and robustness to faults and abuse. While the study of consensus has a 40+ year tradition, the new Internet-scale setting requires a fundamental rethinking of models, desiderata, and protocols. An emergent key challenge is to simultaneously serve clients with different requirements regarding the two fundamental aspects of security, liveness ("good things happen") and safety ("bad things don't happen"). For different instances of this theme, I present the first protocols that allow optimal liveness-safety tradeoff. Results from this line of work have found adoption in the Ethereum blockchain that powers an ecosystem worth $500bn+.Short BioJoachim Neu is a PhD candidate at Stanford University advised by David Tse. His research focuses on Internet-scale consensus as a key enabler for decentralized systems, and spans distributed systems, probabilistic systems analysis, applied cryptography, and networking/communications. Website:
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 601
UID:eventx6a5a287eebb0610516
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20240117T121500
DTEND;TZID=Asia/Jerusalem:20240117T131500
DTSTAMP;TZID=Asia/Jerusalem:20240117T121500
SUMMARY: Theory Seminar  talk by David Wajc (Technion)  Theory Seminar: Online edge coloring  at 2024-01-17 12:15:00
DESCRIPTION:Vizing’s Theorem provides an algorithm that edge colors any graph of maximum degree Δ using Δ+1 colors, which is necessary for some graphs, and at most one higher than necessary for any graph. In online settings, the trivial greedy algorithm requires 2Δ-1 colors, and Bar-Noy, Motwani and Naor in the early 90s showed that this is best possible, at least in the low-degree regime. In contrast, they conjectured that for graphs of superlogarithmic-in-n maximum degree, much better can be done, and that even (1+o(1))Δ-colors suffice online. In this talk I will outline the history of this conjecture, and its recent resolution, together with extensions of a flavor resembling classic and recent results on *list* edge-coloring and “local” edge-coloring.\n\nTalk based in part on joint works with many wonderful and colorful collaborators, including Sayan Bhattacharya, Joakim Blikstad, Ilan R. Cohen, Fabrizio Grandoni, Seffi Naor, Binghui Peng, Amin Saberi, Aravind Srinivasan, Ola Svensson and Radu Vintan.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 201
UID:eventx6a5a287eebb1810518
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20240118T113000
DTEND;TZID=Asia/Jerusalem:20240118T123000
DTSTAMP;TZID=Asia/Jerusalem:20240118T113000
SUMMARY: cggc  talk by Prof. Misha Kazhdan (Computer Science, Johns Hopkins University)  CGGC Seminar: Poisson Manifold Reconstruction (beyond co-dimension one)  at 2024-01-18 11:30:00
DESCRIPTION:In this talk we consider the problem of manifold reconstruction from oriented point clouds for embedded manifolds of co-dimension larger than one. Using the framework of Poisson Surface Reconstruction, and formulating the problem in the language of alternating products, we show that the earlier approach for reconstructing hyper-surfaces extends to general manifolds, at the cost of replacing a quadratic energy with a multi-quadratic energy. We provide an efficient iterative hierarchical solver that empirically converges to a good reconstruction. We show examples reconstructing curves in 3D and surfaces in 4D. And we validate that the approach remains stable in the presence of sampling and noise.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub  012 (Learning Center Auditorium)
UID:eventx6a5a287eebb2710513
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20240123T183000
DTEND;TZID=Asia/Jerusalem:20240123T203000
DTSTAMP;TZID=Asia/Jerusalem:20240123T183000
SUMMARY: CSpecial Event  The Technion CTF Team opens at the Faculty  at 2024-01-23 18:30:00
DESCRIPTION:The Technion CTF Team opens at the Faculty\n\nWe invite you to join the Capture The Flag - CTF meetings. CTF is a cyber challenge competition and information security on the topics: cryptography, reverse engineering, forensics, web, etc. The meetings will include guest lectures and practical experience in solving challenges.\nBeginner and experienced students are welcome to join.\n\nThe first introductory meeting will be held on Tuesday, January 23 at 6:30 pm at Taub 9.\nIn the future, face-to-face meetings will be held every Tuesday at 6:30 pm in Taub.\n\nRegister at the link (the number of places is limited)\n
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 9
UID:eventx6a5a287eebb3610525
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20240124T121500
DTEND;TZID=Asia/Jerusalem:20240124T131500
DTSTAMP;TZID=Asia/Jerusalem:20240124T121500
SUMMARY: Theory Seminar  talk by Daniel Carmon (Technion)  Theory Seminar: The Sample Complexity Of ERMs In Stochastic Convex Optimization  at 2024-01-24 12:15:00
DESCRIPTION:Stochastic convex optimization is one of the most well-studied models for learning in modern machine learning. Nevertheless, a central fundamental question in this setup remained unresolved:\nHow many data points must be observed so that any empirical risk minimizer (ERM) shows good performance on the true population?\nThis question was proposed by Feldman who proved that Ω(\frac{d}{ϵ} + \frac{1}{ϵ^2}) data points are necessary (where d is the dimension and ε > 0 is the accuracy parameter). Proving an ω(\frac{d}{ϵ} + \frac{1}{ϵ^2} ) lower bound was left as an open problem. In this work we show that in fact \tilde{O}(\frac{d}{ϵ} + \frac{1}{ϵ^2}) data points are also sufficient. This settles the question and yields a new separation between ERMs and uniform convergence. This sample complexity holds for the classical setup of learning bounded convex Lipschitz functions over the Euclidean unit ball. We further generalize the result and show that a similar upper bound holds for all symmetric convex bodies. The general bound is composed of two terms: (i) a term of the form \tilde{O}(\frac{d}{ϵ}) with an inverse-linear dependence on the accuracy parameter, and (ii) a term that depends on the statistical complexity of the class of linear functions (captured by the Rademacher complexity). The proof builds a mechanism for controlling the behavior of stochastic convex optimization problems.\n
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 201
UID:eventx6a5a287eebb4510524
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20240128T103000
DTEND;TZID=Asia/Jerusalem:20240128T113000
DTSTAMP;TZID=Asia/Jerusalem:20240128T103000
SUMMARY: colloq  talk by Hadar Frenkel   CS Lecture: Verification of Complex Hyperproperties  at 2024-01-28 10:30:00
DESCRIPTION:Hyperproperties are system properties that relate multiple execution traces to one another. Hyperproperties are essential to express a wide range of system requirements such as information flow and security policies; epistemic properties like knowledge in multi-agent systems; fairness; and robustness. With the aim of verifying program correctness, the two major challenges are (1) providing a specification language that can precisely express the desired properties; and (2) providing scalable verification algorithms. In this talk, I will give an overview of my recent work on addressing these two challenges.\n \nFirst, I will present the new logic Hyper^2LTL, which uses second-order quantification over sets of executions to express a wide range of complex hyperproperties such as common knowledge in multi-agent systems and asynchronous hyperproperties.\nSecond, I will present a (sound but necessarily incomplete) model-checking algorithm for Hyper^2LTL; While the verification of Hyper^2LTL is undecidable, we manage to handle undecidability by characterizing a rich fragment of the logic that allows for sound approximations, providing the first verification algorithm for Hyper^2LTL specifications. \nLast, I will outline my work on causal analysis in reactive systems, both in formalizing causality to the setting of reactive systems and infinite executions and in algorithms for verifying and generating explanations.\n\nAffiliation: CISPA Helmholtz Center for Information Security.\n\nShort bio:\nHadar Frenkel is a postdoctoral researcher at CISPA Helmholtz Center for Information Security in Saarbrücken, Germany, hosted by Prof. Bernd Finkbeiner. She obtained her PhD from the Technion in 2021, under the supervision of Prof. Orna Grumberg and Dr. Sarai Sheinvald. Hadar studies complex hyperproperties, such as knowledge, causality, and asynchronous hyperproperties, and searches for logical formalisms and verification algorithms for them. She also studies automata learning and its applications in program verification and repair.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Auditorium 012, floor 0
UID:eventx6a5a287eebb5410522
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20240129T153000
DTEND;TZID=Asia/Jerusalem:20240129T163000
DTSTAMP;TZID=Asia/Jerusalem:20240129T153000
SUMMARY: MSC  talk by Michael Toker  Diffusion Lens: Interpreting Text Encoders in Text-to-Image Pipelines  at 2024-01-29 15:30:00
DESCRIPTION:Text-to-image diffusion models (T2I) use a latent representation of a text prompt to guide the image generation process.\nHowever, the encoder that produces the text representation is largely unexplored. We propose the Diffusion Lens, a method for analyzing the text encoder of T2I models by generating images from its intermediate representations. Using the Diffusion Lens, we perform an extensive analysis of two recent T2I models.We find that the text encoder gradually builds prompt representations across multiple scenarios.\nComplex scenes describing multiple objects are composed progressively and more slowly than simple scenes; earlier layers encode the concepts in the prompts without a clear interaction, which emerges only in later layers. Moreover, the retrieval of uncommon concepts requires further computation until a faithful representation of the prompt is achieved. Concepts are built from coarse to fine, with details being added until the very late layers. Overall, our findings provide valuable insights into the text encoder component in T2I pipelines.\n\n
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 601
UID:eventx6a5a287eebb6610519
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20240131T103000
DTEND;TZID=Asia/Jerusalem:20240131T113000
DTSTAMP;TZID=Asia/Jerusalem:20240131T103000
SUMMARY: colloq  talk by Gal Vardi  CS Lecture: On Implicit Bias and Benign Overfitting in Neural Networks  at 2024-01-31 10:30:00
DESCRIPTION:When training large neural networks, there are typically many solutions that perfectly fit the training data. Nevertheless, gradient-based methods often have a tendency to reach those which generalize well, namely, perform well also on test data. Thus, the training algorithm seems to be implicitly biased towards certain networks, which exhibit good generalization performance. Understanding this “implicit bias” has been a subject of extensive research recently. Moreover, in contradiction to conventional wisdom in machine learning theory, trained networks often generalize well even when perfectly fitting noisy training data (i.e., data with label noise), a phenomenon called “benign overfitting”.In this talk, I will discuss the above phenomena. In the first part of the talk, I will discuss the implicit bias and its implications. I will show how the implicit bias can lead to good generalization performance, but can also have negative implications in the context of susceptibility to adversarial examples and privacy attacks. In the second part of the talk, I will explore benign overfitting and the settings in which it occurs in neural networks.Short bio:Gal is a postdoctoral researcher at TTI-Chicago and the Hebrew University, hosted by Nati Srebro and Amit Daniely as part of the NSF/Simons Collaboration on the Theoretical Foundations of Deep Learning. Prior to that, he was a postdoc at the Weizmann Institute, hosted by Ohad Shamir, and a PhD student at the Hebrew University, advised by Orna Kupferman. His research focuses on theoretical machine learning, with an emphasis on deep-learning theory.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 601
UID:eventx6a5a287eebb7510528
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20240131T113000
DTEND;TZID=Asia/Jerusalem:20240131T123000
DTSTAMP;TZID=Asia/Jerusalem:20240131T113000
SUMMARY: ceClub  talk by Dr. Yaniv David (Columbia University)  ceClub: Challenges and Opportunities In Securing Software Supply Chains  at 2024-01-31 11:30:00
DESCRIPTION:Racing to be first to market and deploy new features, developers rely on many external libraries to underpin their software. Each library uses more libraries, creating vast networks of dependencies that the developers know little about and have no control over, forming a knowledge gap that quickly turns into technical debt. Repaying this debt is difficult, as analyzing or examining all libraries is infeasible, and worse, the debt keeps growing due to frequent library updates. Attackers move quickly to collect on this debt by reverse-engineering security updates into 1-day attacks or injecting malicious code into libraries.In this talk I will present the systems I built to tackle these challenges: (1) detecting vulnerable libraries in firmware by comparing multiple significant code segments aligned via re-optimizing and normalizing; (2) streamlining software dependency updates via a production-ready hybrid static-dynamic approach for studying the risks of the update before applying it; (3) detecting rogue updates via trust-domain-based tracking for data-flows between different packages in JavaScript code; and (4) hardening applications against data deserialization attacks via a novel type inference technique we call Static Duck Typing, which is based on object behaviors and usage.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Meyer 861
UID:eventx6a5a287eebb8510527
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20240131T121500
DTEND;TZID=Asia/Jerusalem:20240131T131500
DTSTAMP;TZID=Asia/Jerusalem:20240131T121500
SUMMARY: Theory Seminar  talk by Dor Katzelnick (Technion)  Theory Seminar: Almost Logarithmic Approximation for Cutwidth and Pathwidth  at 2024-01-31 12:15:00
DESCRIPTION:We study several graph layout problems with a min max objective. Here, given a graph we wish to find a linear ordering of the vertices that minimizes some worst case objective over the natural cuts in the ordering; which separate an initial segment of the vertices from the rest. A prototypical problem here is cutwidth, where we want to minimize the maximum number of edges crossing a cut. The only known algorithm here is by [Leighton-Rao J.ACM 99] based on recursively partitioning the graph using balanced cuts. This achieves an O(log^(3/2)n) approximation using the O(log^(1/2)n) approximation of [Arora-Rao-Vazirani J.ACM 09] for balanced cuts.We depart from the above approach and give an improved O(log^(1+o(1))n) approximation for cutwidth. Our approach also gives a similarly improved O(log^(1+o(1))n) approximation for finding the pathwidth of a graph. Previously, the best known approximation for pathwidth was O(log^(3/2)n).Talk is based on a joint work with Nikhil Bansal and Roy Schwartz.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 201
UID:eventx6a5a287eebb9510529
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20240131T123000
DTEND;TZID=Asia/Jerusalem:20240131T133000
DTSTAMP;TZID=Asia/Jerusalem:20240131T123000
SUMMARY: PHD  talk by Dor Katzelnick  Almost Logarithmic Approximation for Cutwidth and Pathwidth  at 2024-01-31 12:30:00
DESCRIPTION:We study several graph layout problems with a min max objective. Here, given a graph we wish to find a linear ordering of the vertices that minimizes some worst case objective over the natural cuts in the ordering; which separate an initial segment of the vertices from the rest. A prototypical problem here is cutwidth, where we want to minimize the maximum number of edges crossing a cut. The only known algorithm here is by [Leighton-Rao J.ACM 99] based on recursively partitioning the graph using balanced cuts. This achieves an O(log^(3/2)n) approximation using the O(log^(1/2)n) approximation of [Arora-Rao-Vazirani J.ACM 09] for balanced cuts.\nWe depart from the above approach and give an improved O(log^(1+o(1))n) approximation for cutwidth. Our approach also gives a similarly improved O(log^(1+o(1))n) approximation for finding the pathwidth of a graph. Previously, the best known approximation for pathwidth was O(log^(3/2)n).Talk is based on a joint work with Nikhil Bansal and Roy Schwartz, and held together with the theory seminar.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 201
UID:eventx6a5a287eebba410523
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20240201T113000
DTEND;TZID=Asia/Jerusalem:20240201T123000
DTSTAMP;TZID=Asia/Jerusalem:20240201T113000
SUMMARY: Theory Seminar  talk by Gilad Asharov (Bar Ilan University)  Theory Of Crypto Seminar: Perfect Asynchronous MPC with Linear Communication Overhead  at 2024-02-01 11:30:00
DESCRIPTION:We study secure multiparty computation in the asynchronous setting with perfect security and optimal resilience (less than one-fourth of the participants are malicious). It has been shown that every function can be computed in this model [Ben-OR, Canetti, and Goldreich, STOC'1993]. \nDespite 30 years of research, all protocols in the asynchronous setting require $\Omega(n^2C)$ communication complexity for computing a circuit with $C$ multiplication gates. In contrast, for nearly 15 years, in the synchronous setting, it has been known how to achieve $O(nC)$ communication complexity (Beerliova and Hirt; TCC 2008). The techniques for achieving this result in the synchronous setting are not known to be sufficient for obtaining an analogous result in the asynchronous setting.We close this gap between synchronous and asynchronous secure computation and show the first asynchronous protocol with $O(nC)$ communication complexity for a circuit with $C$ multiplication gates. Linear overhead forms a natural barrier for general secret-sharing-based MPC protocols. Our main technical contribution is an asynchronous weak binding secret sharing that achieves rate-1 communication (i.e., $O(1)$-overhead per secret). To achieve this goal, we develop new techniques for the asynchronous setting, including the use of \emph{trivariate polynomials} (as opposed to bivariate polynomials).Joint work with Ittai Abraham, Shravani Patil, and Arpita Patra\n
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 401
UID:eventx6a5a287eebbb410526
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20240206T113000
DTEND;TZID=Asia/Jerusalem:20240206T123000
DTSTAMP;TZID=Asia/Jerusalem:20240206T113000
SUMMARY: pixel-club  talk by Guy Gaziv (DiCarlo Lab at MIT)  Pixel-Club: Robustified ANNs Reveal Wormholes Between Human Category Percepts  at 2024-02-06 11:30:00
DESCRIPTION:The visual object category reports of artificial neural networks (ANNs) are notoriously sensitive to tiny, adversarial image perturbations. Because human category reports (aka human percepts) are thought to be insensitive to those same small-norm perturbations — and locally stable in general — this argues that ANNs are incomplete scientific models of human visual perception. Consistent with this, we show that when small-norm image perturbations are generated by standard ANN models, human object category percepts are indeed highly stable. However, in this very same "human-presumed-stable" regime, we find that robustified ANNs reliably discover low-norm image perturbations that strongly disrupt human percepts. These previously undetectable human perceptual disruptions are massive in amplitude, approaching the same level of sensitivity seen in robustified ANNs. Further, we show that robustified ANNs support precise perceptual state interventions: they guide the construction of low-norm image perturbations that strongly alter human category percepts toward specific prescribed percepts. These observations suggest that for arbitrary starting points in image space, there exists a set of nearby "wormholes", each leading the subject from their current category perceptual state into a semantically very different state. Moreover, contemporary ANN models of biological visual processing are now accurate enough to consistently guide us to those portals.Short Bio:Guy is a Computer Vision postdoctoral researcher at the DiCarlo Lab at MIT, interested in the intersection between machine and human vision. His PhD focused on decoding visual experience from brain activity. His current focus is on harnessing contemporary models of primate visual cognition for neural and behavioral modulation. Guy holds a PhD in Computer Science and an MSc in Physics from The Weizmann Institute of Science, and a BSc in Physics-EECS from The Hebrew University of Jerusalem.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Room 1061, EE Meyer Building 
UID:eventx6a5a287eebbc410532
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20240206T123000
DTEND;TZID=Asia/Jerusalem:20240206T133000
DTSTAMP;TZID=Asia/Jerusalem:20240206T123000
SUMMARY: MSC  talk by Elad Kinsbruner  Constrictor: Immutability as a Design Concept  at 2024-02-06 12:30:00
DESCRIPTION:Many object-oriented applications in algorithm design rely on objects never changing during their lifetime. This is often tackled by marking object references as read-only, e.g., using the const keyword in C++. In other languages like Python or Java where such a concept does not exist, programmers rely on best practices that are entirely unenforced. While reliance on best practices is obviously too permissive, const-checking is too restrictive: it is possible for a method to mutate the internal state while still satisfying the property we expect from an “immutable” object in this setting. We would therefore like to enforce the immutability of an object’s abstract state.We check an object’s immutability through a view of its abstract state: for instances of an immutable class, the view does not change when running any of the class’s methods, even if some of the internal state does change. If all methods of a class are verified as non-mutating, we can deem the entire class view-immutable. We present an SMT-based algorithm to check view-immutability, and implement it in our linter, Constrictor.We evaluate Constrictor on 52 examples of immutability-related design violations. Our evaluation shows that Constrictor is effective at catching a variety of prototypical design violations, and does so in seconds. We also explore Constrictor with several case studies.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 601
UID:eventx6a5a287eebbd610520
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20240206T183000
DTEND;TZID=Asia/Jerusalem:20240206T203000
DTSTAMP;TZID=Asia/Jerusalem:20240206T183000
SUMMARY: CSpecial Event  Technion CTF Team  at 2024-02-06 18:30:00
DESCRIPTION:Come be part of a new Capture The Flag-CTF group at the Faculty\n\nThe meetings are held every Tuesday at 18:30 at Taub 9 and include guest lectures and practical experience in solving challenges.\nEverybody is invited! Beginners and experienced\nFor details: Technionctf.com\n\nThe week of February 6, 2024:\nBeginners: Forensics &amp; Networks | Taub 9\nexperienced: Challenges from 2023 LA CTF | Taub 8
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 9 
UID:eventx6a5a287eebbe710534
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20240207T121500
DTEND;TZID=Asia/Jerusalem:20240207T131500
DTSTAMP;TZID=Asia/Jerusalem:20240207T121500
SUMMARY: Theory Seminar  talk by Konstantin Zabaranyi (Technion)  Theory Seminar: Algorithmic Cheap Talk  at 2024-02-07 12:15:00
DESCRIPTION:Come be part of a new Capture The Flag-CTF group at the Faculty\n\nThe meetings are held every Tuesday at 18:30 at Taub 9 and include guest lectures and practical experience in solving challenges.\n\nEverybody is invited! Beginners and experienced\nFor details: Technionctf.com\n\nThe week of February 6, 2024:\n\nBeginners: Forensics &amp; Networks | Taub 9\n\nexperienced: Challenges from 2023 LA CTF | Taub 8
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 201
UID:eventx6a5a287eebbf710533
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20240207T123000
DTEND;TZID=Asia/Jerusalem:20240207T143000
DTSTAMP;TZID=Asia/Jerusalem:20240207T123000
SUMMARY: CSpecial Event  Final Spotlight Day  at 2024-02-07 12:30:00
DESCRIPTION:You are invited to Final spotlight day \nWednesday 07.02.2024 | 12:30-14:30 | Visitor Center Auditorium 012, Floor 0\n\n12:30 - Come meet engineers, researchers and the recruitment team at Final, and get to know the day-to-day life at Final.\n\n13:15 - Meeting on options, probabilities and the world of algorithm trading | Noam Horowitz - researcher at Final\n\nTo register for the lecture click here (the number of places is limited).
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Auditorium 012, floor 0
UID:eventx6a5a287eebc0610531
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20240212T120000
DTEND;TZID=Asia/Jerusalem:20240212T130000
DTSTAMP;TZID=Asia/Jerusalem:20240212T120000
SUMMARY: MSC  talk by Robert Shahla  A 0-RTT-Aware QUIC Load Balancer  at 2024-02-12 12:00:00
DESCRIPTION:QUIC is an emerging transport protocol, offering multiple advantages over TCP. Yet, to fully unleash QUIC’s potential, a paradigm shift is needed in existing network infrastructure. We propose a novel 0-RTT-aware load balancing algorithm. 0-RTT is crucial for web performance, particularly on mobile networks. Our load balancing algorithm ensures 0-RTT while maintaining near-optimal load balancing performance.Through extensive simulations, using both synthetic and real-world traffic traces, we demonstrate that the proposed load balancer achieves near-optimal load balancing performance, faster time-to-first-byte, and faster completion time than the least loaded, power-of-two-choices, and maximum session affinity load balancing policies.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture: 97318309550 and Taub 601
UID:eventx6a5a287eebc1610530
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20240213T183000
DTEND;TZID=Asia/Jerusalem:20240213T203000
DTSTAMP;TZID=Asia/Jerusalem:20240213T183000
SUMMARY: CSpecial Event  Technion CTF Team  at 2024-02-13 18:30:00
DESCRIPTION:Come be part of a new Capture The Flag-CTF group at the Faculty\n\nThe meetings are held every Tuesday at 18:30 at Taub 9 and include guest lectures and practical experience in solving challenges.\nEverybody is invited! Beginners and experienced\nFor details: Technionctf.com\n\nThis week - February 13, 2024:\n\n18:30 | Taub 2 | Omar Atias - security researcher, lecturer at BlackHat USA & DEFCON\nThe price of convenience- how weaknesses in payment systems of transportation services can cost you dearly\n\n19:15 | Practice challenges from CTFs\n
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 2 
UID:eventx6a5a287eebc2510536
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20240214T113000
DTEND;TZID=Asia/Jerusalem:20240214T123000
DTSTAMP;TZID=Asia/Jerusalem:20240214T113000
SUMMARY: ceClub  talk by Ron Marcus   ceClub: Space-efficient FTL for Mobile Storage via Tiny Neural Nets  at 2024-02-14 11:30:00
DESCRIPTION:With the rapid increase of storage demands and working sets of modern mobile apps, maintaining high I/O performance in mobile SSDs under strict resource constraints is challenging. The Flash Translation Layer (FTL) must increase the capacity of the Logical-To-Physical (L2P) address translation cache to keep up with the new workloads, but it comes at the cost of scaling the on-die SRAM, resulting in higher chip area, power consumption, and costs.In this talk, I will present RQFTL, a demand-based FTL for mobile storage controllers that boosts the effective cache capacity over state-of-the-art techniques. RQFTL stores a large part of the L2P cache in a compressed form, and employs a learned data structure called RQRMI that leverages tiny neural nets to quickly find the correct translation entry in the cache. RQFTL uses neural network inference for cache lookups, and rapidly retrains the neural nets to efficiently handle L2P cache updates. It is specifically optimized to achieve high coverage for scattered read accesses, making it suitable for popular read-skewed workloads such as mobile gaming.The talk includes an evaluation of RQFTL on hours-long real-world I/O traces of popular modern mobile apps including games, video editing and social networking apps collected on Google Pixel V6 Phone. It shows that RQFTL outperforms all the state-of-the-art FTLs in these workloads, increasing the effective L2P cache capacity by over an order of magnitude compared to  DFTL and up to 5X over the recent LeaFTL.   As a result, it achieves 2X and 1.42X  higher hit rate compared to DFTL and LeaFTL respectively, under the same SRAM capacity, and allows reduction of the total SRAM capacity of a controller by about a third of that of LeaFTL.Ron is an MSc student supervised by Prof. Mark Silberstein.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Meyer 861
UID:eventx6a5a287eebc3310535
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20240214T121500
DTEND;TZID=Asia/Jerusalem:20240214T131500
DTSTAMP;TZID=Asia/Jerusalem:20240214T121500
SUMMARY: Theory Seminar  talk by Idan Mehalel (Technion)  Theory Seminar: Optimal Prediction Using Expert Advice and Randomized Littlestone Dimension  at 2024-02-14 12:15:00
DESCRIPTION:Suppose that n forecasting experts (or functions) are providing daily rain/no-rain predictions, and the best among them is mistaken in at most k many days. For how many days will a person allowed to observe the predictions, but has no weather-forecasting prior knowledge, mispredict?In this talk, we will discuss how such classical problems can be reduced to calculating the (average) depth of binary trees, by using newly introduced complexity measures (aka dimensions) of the set of experts/functions. All of those measures are variations of the classical Littlestone dimension (Littlestone ’88).For the forecasting problem outlined above, Cesa-Bianchi, Freund, Helmbold, and Warmuth [’93, ’96] provided a nearly optimal bound for deterministic learners, and left the randomized case as an open problem. We resolve this problem by providing an optimal randomized learner, and showing that its expected mistake bound equals half of the deterministic bound of Cesa-Bianchi et al., up to negligible additive terms.For general (possibly infinite) function classes, we show that the optimal expected regret (= #mistakes – k) when learning a function class with Littlestone dimension d is of order d + (kd)^0.5.Based on a joint work with Yuval Filmus, Steve Hanneke, and Shay Moran.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 201
UID:eventx6a5a287eebc4410538
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20240215T110000
DTEND;TZID=Asia/Jerusalem:20240215T120000
DTSTAMP;TZID=Asia/Jerusalem:20240215T110000
SUMMARY: MSC  talk by Idan Levy  A Deep Learning Platform for Diagnosing ECG Tests  at 2024-02-15 11:00:00
DESCRIPTION:In clinical settings, a significant portion of ECG data is typically available in printed form, and the most convenient means of digitizing this information involves utilizing a mobile device. Despite notable progress in AI-based techniques for paper-based 12-lead ECG analysis, their adoption in clinical practice remains limited primarily due to challenges such as inadequate accuracy in clinical settings and a restricted ability to diagnose various cardiac conditions.\nOur objective was to tackle these challenges through BeatBox AI, an automated diagnostic platform for 12-lead ECGs images. This innovative platform is crafted as a continually self-optimizing solution, offering scalability and adaptability to diverse populations and various 12-lead ECG layouts.\nBeatBox AI's diagnostic capabilities were assessed using 4,117 12-lead ECGs images gathered during a 10-month prospective clinical study, between March 29th, 2023 to January 10th, 2023. This study involved collaboration with 14 cardiologists from five distinct centers located in Israel, Japan, Italy, and Russia. Additionally, publicly accessible online platforms containing 12-lead ECG images and their corresponding diagnosis were utilized in the evaluation process. The study included participants who were random individuals who visited the centers and data uploaded to blogs.The main outcomes in the comparative analysis between the automatic diagnosis and the cardiologists' assessments were that the platform demonstrated improvement across all operational dimensions.The platform could initially diagnose 21 cardiac conditions with an average Matthews Correlation Coefficient (MCC) of 0.19. However, at the concluding evaluation stage, the MCC significantly improved to 0.57, and the platform successfully identified a total of 54 distinct cardiac conditions.The data-driven strategy used in this study allows the platform to enhance its diagnostic accuracy continually. It enables the expansion of the range of diagnosable conditions beyond critical clinical thresholds, even with a small number of clinical samples. Additionally, the platform can adapt seamlessly to newly encountered ECG layouts, showcasing its flexibility and robust learning capabilities.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture: Link and Taub 401
UID:eventx6a5a287eebc5510539
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20240219T133000
DTEND;TZID=Asia/Jerusalem:20240219T143000
DTSTAMP;TZID=Asia/Jerusalem:20240219T133000
SUMMARY: TDC Seminar  talk by Yuval Gil, Technion  TDC Seminar: Deterministic Distributed Maximum Weight Independent Set Approximation in Sparse Graphs  at 2024-02-19 13:30:00
DESCRIPTION:This talk focuses on the distributed task of constructing an approximate \emph{maximum weight independent set (MWIS)}.\nSpecifically, we are interested in deterministic CONGEST algorithms whose approximation guarantees are expressed as a function of the graph's \emph{arboricity} $\alpha$.Generally speaking, efficient deterministic non-trivial approximation algorithms for MWIS were not known until the recent breakthrough of Faour et al.~[SODA 2023] that obtained an $O(\Delta)$-approximation in $O(\log^{2} n)$ rounds on graphs of maximum degree $\Delta$. Combined with a transformation presented by Kawarabayashi et al.~[DISC 2020], one obtains the current state-of-the-art: a $4 (1 + \epsilon) \alpha$-approximation in $O(\log^{3} n)$ rounds. In this talk, I present new algorithms that achieve arboricity-dependent approximations for MWIS.These algorithms exhibit an approximation-runtime tradeoff: on one endpoint of the spectrum is an improved $(2 + \epsilon) \alpha$ approximation in $O(\alpha \log n)$ rounds; on the other endpoint is an $\alpha^{1 + \epsilon}$-approximation with a significantly faster $O(\log \alpha \log n)$ runtime.The talk will be self contained.\n\n\n
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zisapel (ECE) 608
UID:eventx6a5a287eebc6810541
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20240221T121500
DTEND;TZID=Asia/Jerusalem:20240221T131500
DTSTAMP;TZID=Asia/Jerusalem:20240221T121500
SUMMARY: Theory Seminar  talk by Gal Arnon (Weizmann institute)  Theory Seminar: IOPs with Inverse Polynomial Soundness Error  at 2024-02-21 12:15:00
DESCRIPTION:We show that every language in NP has an Interactive Oracle Proof (IOP) with inverse polynomial soundness error and small query complexity. This achieves parameters that surpass all previously known PCPs and IOPs. Specifically, we construct an IOP with perfect completeness, soundness error 1/n, round complexity O(loglog n), proof length poly(n) over an alphabet of size O(n), and query complexity O(loglog n). This is a step forward in the quest to establish the sliding-scale conjecture for IOPs (which would additionally require query complexity O(1)).Our main technical contribution is a emph{high-soundness small-query} proximity test for the Reed–Solomon code. We construct an IOP of proximity for Reed–Solomon codes, over a field F with evaluation domain L and degree d, with perfect completeness, soundness error (roughly) max{1-delta , O(rho^{1/4})}$ for delta-far functions, round complexity O(loglog d), proof length O(|L|/rho) over F, and query complexity O(loglog d); here rho = (d+1)/|L| is the code rate. En route, we obtain a new high-soundness proximity test for bivariate Reed–Muller codes.The IOP for NP is then obtained via a high-soundness reduction from NP to Reed–Solomon proximity testing with rate rho = 1/poly(n) and distance delta = 1-1/poly(n) (and applying our proximity test). Our constructions are direct and efficient, and hold the potential for practical realizations that would improve the state-of-the-art in real-world applications of IOPs.Based on joint work with Alessandro Chiesa and Eylon Yogev
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 201
UID:eventx6a5a287eebc7a10544
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20240221T150000
DTEND;TZID=Asia/Jerusalem:20240221T160000
DTSTAMP;TZID=Asia/Jerusalem:20240221T150000
SUMMARY: colloq  talk by Shira Faigenbaum-Golovin  Identifying Underlying Geometry To Analyze High-Dimensional Data: Images And Shape Spaces As A Case Study  at 2024-02-21 15:00:00
DESCRIPTION:High-dimensional data is increasingly available in diverse applications, ranging from images to shapes represented as point clouds. Such data raises novel questions and offers a unique opportunity to study them by developing new machine-learning tools. While the analysis of an individual sample may be challenging, leveraging the power of the data collection can be effective in tackling complex tasks. This talk delves into the challenges associated with studying such data, particularly focusing on learning from limited, non-labeled, noisy data in high dimensions. The discussion centers on the assumption that high-dimensional data represents an embedding of a low-dimensional manifold.First, I will introduce a rigorous framework designed for denoising and reconstructing a low-dimensional manifold in a high-dimensional space from scattered data, by solving a non-convex optimization problem. Next, we will examine scatter data that has its own geometry (i.e. deal with manifold of manifolds), and address questions related to shape registration and variation within the realm of shape spaces. I will demonstrate the methodology on manifolds of various dimensions, as well as on a collection of anatomical surfaces pertaining to evolutionary anthropology. Lastly, I will underscore the importance of manifold learning in the realm of image processing. I will illustrate this through a novel method for comparing handwriting, which has revolutionized the study of ancient inscriptions (published in PNAS).Short Bio: Shira Faigenbaum-Golovin is an Assistant Research Professor at Duke University, working with Ingrid Daubechies. Currently, her work ranges between understanding the theoretical properties of Neural networks and developing computational algorithms to address questions in shape space. Shira received her B.S.c in Mathematics with a major in Computer Science, as well as an M.Sc (in 2014, Magna Cum Laude) and Ph.D. (2021) in Applied Mathematics all from Tel-Aviv University. In parallel to her Ph.D. Shira held an algorithmic expert position in the Image Signal Processor team at Intel. Shira is a recipient of the 2016 Tel-Aviv Dean’s Excellence Scholarship, the 2017 Minerva Research Grant, and the 2021-2023 Zuckerman Postdoctoral Fellowship, as well as has also received generous support from the Schmidt Postdoctoral Award for women in mathematical and computing sciences during 2021-2023. Her research spans from low and high-dimensional approximation, theoretical and applicative machine learning, data science, and image processing.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture:93613579310 Passcode: 652820
UID:eventx6a5a287eebc8d10545
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20240221T170000
DTEND;TZID=Asia/Jerusalem:20240221T180000
DTSTAMP;TZID=Asia/Jerusalem:20240221T170000
SUMMARY: PHD  talk by Daniella Bar-Lev  Theory and Practice of DNA Storage  at 2024-02-21 17:00:00
DESCRIPTION:In the last decade, DNA-based storage systems emerged as a potential data archival solution due to their high data density and durability. This research delves into intrinsic error characteristics of DNA storage systems to devise robust coding strategies and innovative algorithms for enhanced reliability, efficiency, scalability, and cost-effectiveness. The research propels DNA storage feasibility while contributing to foundational theory.The work analyzes combinatorial structures tied to errors that are common in DNA storage, like insertions and deletions, alongside examining corresponding channels. These explorations illuminate DNA storage's unique error behavior, offering insights into its fundamental limits and capabilities.Moreover, to advance DNA storage-related coding techniques, this work introduces novel code constructions spanning from innovative reconstruction codes to streamlined constraint coding algorithms. These efforts underscore coding techniques' pivotal role in DNA storage advancement.Bridging theory with application, the DNA coverage depth problem, a delicate balance involving sequencing costs, latency, and retrieval accuracy, is introduced and addressed in this work. A pioneering proof-of-concept for a scalable DNA storage pipeline that integrates Deep Neural Networks with coding strategies is introduced. It enables error-free retrieval with state-of-the-art accuracy and improvement of orders of magnitude in efficiency.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture: 92916633492 and Taub 9
UID:eventx6a5a287eebca110540
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20240222T150000
DTEND;TZID=Asia/Jerusalem:20240222T160000
DTSTAMP;TZID=Asia/Jerusalem:20240222T150000
SUMMARY: colloq  talk by Tomer Galanti   Fundamental Problems in AI: Transferability, Compressibility and Generalization  at 2024-02-22 15:00:00
DESCRIPTION:In this talk, we delve into several fundamental questions in deep learning. We start by addressing the question, "What are good representations of data?" Recent studies have shown that the representations learned by a single classifier over multiple classes can be easily adapted to new classes with very few samples. We offer a compelling explanation for this behavior by drawing a relationship between transferability and an emergent property known as neural collapse. Later, we explore why certain architectures, such as convolutional networks, outperform fully-connected networks, providing theoretical support for how their inherent sparsity aids learning with fewer samples. Lastly, I present recent findings on how training hyperparameters implicitly control the ranks of weight matrices, consequently affecting the model's compressibility and the dimensionality of the learned features.Additionally, I will describe how this research integrates into a broader research program where I aim to develop realistic models of contemporary learning settings to guide practices in deep learning and artificial intelligence. Utilizing both theory and experiments, I study fundamental questions in the field of deep learning, including why certain architectural choices improve performance or convergence rates, when transfer learning and self-supervised learning work, and what kinds of data representations are learned in practical settings.Short Bio:Tomer Galanti is a Postdoctoral Associate at the Center for Brains, Minds, and Machines at MIT, where he focuses on the theoretical and algorithmic aspects of deep learning. He received his Ph.D. in Computer Science from Tel Aviv University and served as a Research Scientist Intern at Google DeepMind's Foundations team during his doctoral studies. He has published numerous papers in top-tier conferences and journals, including NeurIPS, ICML, ICLR, and JMLR. His work, titled "On the Modularity of Hypernetworks," was awarded an oral presentation at NeurIPS 2020.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture:9316005503 Passcode: 272625
UID:eventx6a5a287eebcb310546
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20240226T133000
DTEND;TZID=Asia/Jerusalem:20240226T143000
DTSTAMP;TZID=Asia/Jerusalem:20240226T133000
SUMMARY: TDC Seminar  talk by Naama Ben-David, Technion  TDC Seminar: Fast and Fair Lock-Free Locks  at 2024-02-26 13:30:00
DESCRIPTION:Locks are frequently used in concurrent systems to simplify code and ensure safe access to contended parts of memory. However, they are also known to cause bottlenecks in concurrent code, leading practitioners and theoreticians to sometimes opt for more intricate lock-free implementations. In this talk, I’ll show that, despite the seeming contradiction, it is possible to design practically and theoretically efficient lock-free locks; I'll present a lock-free lock algorithm with good bounds on running time and fairness.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zisapel (ECE) 608
UID:eventx6a5a287eebcc710548
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20240228T121500
DTEND;TZID=Asia/Jerusalem:20240228T131500
DTSTAMP;TZID=Asia/Jerusalem:20240228T121500
SUMMARY: Theory Seminar  talk by Sarel Cohen (Tel-Aviv Academic College)  Distance Sensitivity Oracles  at 2024-02-28 12:15:00
DESCRIPTION:Theory Seminar: An f-edge fault-tolerant distance sensitivity oracle (f-DSO) is a data-structure that, when queried with two vertices (s, t) and a set F of at most f edges of a graph G with n vertices, returns an estimate tilde{d}(s,t,F) of the distance d(s,t,F) from s to t in G – F. The oracle has stretch alpha if the estimate satisfies d(s,t,F) le tilde{d}(s,t,F) le alpha cdot d(s,t,F) . In the last two decades, extensive research has focused on developing efficient f-DSOs. This research aims to optimize preprocessing time, space consumption, and query time, as well as to improve the stretch (approximation guarantee). Efforts have also been made to derandomize the construction and query algorithms of these systems. Over the last two decades, extensive research has been conducted on developing efficient f-DSOs. This research has focused on optimizing various aspects such as preprocessing time, space consumption, query time, and stretch (approximation guarantee). Efforts have also been made to derandomize the construction algorithms of these data-structures. While multiple constructions of f-DSOs are already known, surprisingly, there is still no optimal data-structure that supports multiple failures. In this talk, we will cover several recent f-DSO data-structures and discuss open questions in this field.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 201
UID:eventx6a5a287eebcd810549
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20240228T123000
DTEND;TZID=Asia/Jerusalem:20240228T143000
DTSTAMP;TZID=Asia/Jerusalem:20240228T123000
SUMMARY: CSpecial Event  Spotlight Day For Nvidia At The Technion  at 2024-02-28 12:30:00
DESCRIPTION:Spotlight day for Nvidia at the Technion on February 28, 2024\n\nNvidia is coming to meet the students of the Faculty of Computer Science at the Technion!\nCome and meet the company's engineers face to face\n\nWednesday February 28, 2024 | 12:30-14:30 | Taub lobby, floor 0
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:CS Taub Lobby
UID:eventx6a5a287eebce710547
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20240229T150000
DTEND;TZID=Asia/Jerusalem:20240229T160000
DTSTAMP;TZID=Asia/Jerusalem:20240229T150000
SUMMARY: MSC  talk by Neta Friedman  Stable Tuple Embeddings for Dynamic Databases  at 2024-02-29 15:00:00
DESCRIPTION:We study the problem of computing an embedding of the tuples of a relational database\nin a manner that is extensible to dynamic changes of the database. In this problem,\nthe embedding should be stable in the sense that it should not change on the existing\ntuples due to the embedding of newly inserted tuples (as database applications might\nalready rely on existing embeddings); at the same time, the embedding of all tuples, old\nand new, should retain high quality. This task is challenging since inter-dependencies\namong the embeddings of different entities are inherent in state-of-the-art embedding\ntechniques for structured data.We study two approaches to solving the problem. The first is an adaptation of\nNode2Vec to dynamic databases. The second is the FoRWaRD algorithm (Foreign\nKey Random Walk Embeddings for Relational Databases) that draws from embedding\ntechniques for general graphs and knowledge graphs and is inherently utilizing the\nschema and its key and foreign-key constraints. We evaluate the embedding algorithms\nusing a collection of downstream tasks of column prediction over geographical and\nbiological domains. We find that in the traditional static setting, our two embedding\nmethods achieve comparable results that are compatible with the state-of-the-art for\nthe specific applications. In the dynamic setting, we find that the FoRWaRD algorithm\ngenerally outperforms and runs faster than the alternatives, and moreover, it features\nonly a mild reduction of quality even when the database consists of more than half of\nnewly inserted tuples after the initial training of the embedding.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture: 99751665591
UID:eventx6a5a287eebcf510537
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20240304T183000
DTEND;TZID=Asia/Jerusalem:20240304T203000
DTSTAMP;TZID=Asia/Jerusalem:20240304T183000
SUMMARY: CSpecial Event  talk by Mor Filo, a graduate of the faculty and a developer at Amazon    at 2024-03-04 18:30:00
DESCRIPTION:You have been accepted as a student! What now?\nThe student community at the SHE S faculty invites you to a meeting on:\nFirst time student job: about the opportunities, challenges and skills you acquire in your first job\n\nMonday, 4/3 at 6:30 pm in the piano auditorium\nPlease register in advance: here\nSpeaker: Moore Philo, graduate of the faculty and developer at Amazon\n\nThe meeting is suitable for those who are in the search process and also for those who have started their first job.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Auditorium 012, floor 0
UID:eventx6a5a287eebd0910553
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20240305T150000
DTEND;TZID=Asia/Jerusalem:20240305T160000
DTSTAMP;TZID=Asia/Jerusalem:20240305T150000
SUMMARY: MSC  talk by Liran Ringel  Early Time Classification with Accumulated Accuracy Gap Control  at 2024-03-05 15:00:00
DESCRIPTION:Early time classification algorithms aim to label a stream of features without processing the full input stream, while maintaining accuracy comparable to that achieved by applying the classifier to the entire input. In this paper, we introduce a statistical framework that can be applied to any sequential classifier, formulating a calibrated stopping rule. This data-driven rule attains finite-sample, distribution-free control of the accuracy gap between full and early-time classification. We start by presenting a novel method that builds on the Learn-then-Test calibration framework to control this gap marginally, on average over i.i.d. instances. As this algorithm tends to yield an excessively high accuracy gap for early halt times, our main contribution is the proposal of a framework that controls a stronger notion of error, where the accuracy gap is controlled conditionally on the accumulated halt times. Numerical experiments demonstrate the effectiveness, applicability, and usefulness of our method. We show that our proposed early stopping mechanism reduces up to 94% of timesteps used for classification while achieving rigorous accuracy gap control.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture:96835343656 and Taub 401
UID:eventx6a5a287eebd1410543
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20240306T121500
DTEND;TZID=Asia/Jerusalem:20240306T131500
DTSTAMP;TZID=Asia/Jerusalem:20240306T121500
SUMMARY: Theory Seminar  talk by Yair Carmon (Tel-Aviv university)  Faster Matrix Game Solvers Via Ball Oracle Acceleration  at 2024-03-06 12:15:00
DESCRIPTION:We design a new stochastic first-order algorithm for approximately solving matrix games as well as the more general problem of minimizing the maximum of smooth convex functions. Our central tool is ball oracle acceleration: a technique for minimizing any convex function with a small number of calls to a ball oracle that minimizes the same function restricted to a small ball around the query point. To design an efficient ball oracle for our problems of interest we leverage stochastic gradient descent, softmax smoothing, rejection sampling, and sketching. For a large number of general smooth functions, our algorithm obtains optimal gradient query complexity. For matrix games, it improves over all previous runtime bounds in a range of problem parameters.Based on joint work with Arun Jambulapati, Yujia Jin, and Aaron Sidford.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 201
UID:eventx6a5a287eebd1f10552
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20240306T123000
DTEND;TZID=Asia/Jerusalem:20240306T143000
DTSTAMP;TZID=Asia/Jerusalem:20240306T123000
SUMMARY: CSpecial Event  Cadence Arrive For Recruitment Day At The Faculty  at 2024-03-06 12:30:00
DESCRIPTION:Cadence company is coming to a spotlight day at the faculty\n\nWednesday 6/3 | 12:30-14:30 | Lobby Taub\n\nIn the program: a meeting with the recruitment teams, the engineers for a 1:1 conversation about employment opportunities and tips for writing a report.\nAnd of course merch and sweets.\n\nwaiting for you!!
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:CS Taub Lobby
UID:eventx6a5a287eebd2a10550
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20240310T180000
DTEND;TZID=Asia/Jerusalem:20240310T193000
DTSTAMP;TZID=Asia/Jerusalem:20240310T180000
SUMMARY: CSpecial Event  Nvidia Invites You To a Virtual Spotlight Day 10/3  at 2024-03-10 18:00:00
DESCRIPTION:Nvidia invites you to a virtual spotlight day where the company's engineers will talk about the different groups and open positions\n\nSunday, March 10, from 18:00 to 19:30\n\nTo register for the event here
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Meet virtually
UID:eventx6a5a287eebd3310555
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20240311T103000
DTEND;TZID=Asia/Jerusalem:20240311T113000
DTSTAMP;TZID=Asia/Jerusalem:20240311T103000
SUMMARY: MSC  talk by Zorik Gekhman  On the Robustness of Dialogue History Representation in Conversational Question Answering: A Comprehensive Study and a New Prompt-based Method  at 2024-03-11 10:30:00
DESCRIPTION:Most works on modeling the conversation history in Conversational Question Answering (CQA) report a single main result on a common CQA benchmark. While existing models show impressive results on CQA leaderboards, it remains unclear whether they are robust to shifts in setting (sometimes to more realistic ones), training data size (e.g. from large to small sets) and domain. In this work, we design and conduct the first large-scale robustness study of history modeling approaches for CQA. We find that high benchmark scores do not necessarily translate to strong robustness, and that various methods can perform extremely differently under different settings. Equipped with the insights from our study, we design a novel prompt-based history modeling approach, and demonstrate its strong robustness across various settings. Our approach is inspired by existing methods that highlight historic answers in the passage. However, instead of highlighting by modifying the passage token embeddings, we add textual prompts directly in the passage text. Our approach is simple, easy-to-plug into practically any model, and highly effective, thus we recommend it as a starting point for future model developers. We also hope that our study and insights will raise awareness to the importance of robustness-focused evaluation, in addition to obtaining high leaderboard scores, leading to better CQA systems.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture
UID:eventx6a5a287eebd3e10551
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20240311T133000
DTEND;TZID=Asia/Jerusalem:20240311T143000
DTSTAMP;TZID=Asia/Jerusalem:20240311T133000
SUMMARY: TDC Seminar  talk by Ittai Abraham, Intel  TDC Seminar: From Distributed Computing to Cryptography and Back  at 2024-03-11 13:30:00
DESCRIPTION:I will share some of my learnings from working on problems on the intersection of Distributed Computing and Cryptography.On the one hand, I will show how some cryptographic protocols (MPC and DKG) can be improved by using distributed computing counterparts for notions such as zero knowledge proofs and proofs of knowledge. On the other hand, I will show how distributed computing protocols (ABA) can be improved by carefully using notions of binding from cryptography. One recurring theme is the use of randomization to overcome adaptive malicious adversaries. Another is the use of amortization to overcome lower bounds.Short Bio: Ittai is a senior researcher at Intel Labs and a contributor to https://decentralizedthoughts.github.io. Previously Ittai was at VMware Research and Microsoft Research. He holds a PhD from the Hebrew University.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zisapel (ECE) 608
UID:eventx6a5a287eebd4a10556
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20240312T113000
DTEND;TZID=Asia/Jerusalem:20240312T123000
DTSTAMP;TZID=Asia/Jerusalem:20240312T113000
SUMMARY: pixel-club  talk by Or Avitan (Graduate Seminar)  Pixel-Club: Using Zodiacal Light For Spaceborne Calibration Of Polarimetric Imagers  at 2024-03-12 11:30:00
DESCRIPTION:We propose that spaceborne polarimetric imagers can be calibrated, or self-calibrated using zodiacal light (ZL). ZL is created by a cloud of interplanetary dust particles. It has a significant degree of polarization in a wide field of view. From space, ZL is unaffected by terrestrial disturbances. ZL is insensitive to the camera location, so it is suited for simultaneous cross-calibration of satellite constellations. ZL changes on a scale of months, thus being a quasi-constant target in realistic calibration sessions. We derive a forward model for polarimetric image formation. Based on it, we formulate an inverse problem for polarimetric calibration and self-calibration, as well as an algorithm for the solution. The methods here are demonstrated in simulations. Towards these simulations, we render polarized images of the sky, including ZL from space, polarimetric disturbances, and imaging noise.\nM.Sc. student under the supervision of Prof. Yoav Schechner.\n
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Room 1061, EE Meyer Building 
UID:eventx6a5a287eebd5410561
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20240312T143000
DTEND;TZID=Asia/Jerusalem:20240312T153000
DTSTAMP;TZID=Asia/Jerusalem:20240312T143000
SUMMARY: MSC  talk by Gabriele Serussi  Active Propulsion Noise Shaping For Multi-Rotor Aircraft Localization  at 2024-03-12 14:30:00
DESCRIPTION:Multi-rotor aerial autonomous vehicles (MAVs) primarily rely on vision for navigation purposes. However, visual localization and odometry techniques suffer from poor performance in low or direct sunlight, a limited field of view, and vulnerability to occlusions. Acoustic sensing can serve as a complementary or even alternative modality for vision in many situations, and it also has the added benefits of lower system cost and energy footprint, which is especially important for micro aircraft. This work proposes actively controlling and shaping the aircraft propulsion noise generated by the rotors to benefit localization tasks, rather than considering it a harmful nuisance. We present a neural network architecture for self-noise-based localization in a known environment. We show that training it simultaneously with learning time-varying rotor phase modulation achieves accurate and robust localization. The proposed methods are evaluated using a computationally affordable simulation of MAV rotor noise in 2D acoustic environments that is fitted to real recordings of rotor pressure fields.\n\n
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture:99864029841 and Taub 401
UID:eventx6a5a287eebd7510554
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20240313T103000
DTEND;TZID=Asia/Jerusalem:20240313T113000
DTSTAMP;TZID=Asia/Jerusalem:20240313T103000
SUMMARY: MSC  talk by Guy Bar-Shalom  Window-Based Distribution Shift Detection for Deep Neural Networks‏  at 2024-03-13 10:30:00
DESCRIPTION:To deploy and operate deep neural models in production, the quality of their predictions, which might be contaminated benignly or manipulated maliciously by input distributional deviations, must be monitored and assessed. Specifically, we study the case of monitoring the healthy operation of a deep neural network (DNN) receiving a stream of data, with the aim of detecting input distributional deviations over which the quality of the network’s predictions is potentially damaged. Using selective prediction principles, we propose a distribution deviation detection method for DNNs. The proposed method is derived from a tight coverage generalization bound computed over a sample of instances drawn from the true underlying distribution. Based on this bound, our detector continuously monitors the operation of the network over a test window and fires off an alarm whenever a deviation is detected. Our novel detection method performs on-par or better than the state-of-the-art, while consuming substantially lower computation time (five orders of magnitude reduction) and space complexity. Unlike previous methods, which require at least linear dependence on the size of the source distribution for each detection, rendering them inapplicable to “Google-Scale” datasets, our approach eliminates this dependence, making it suitable for real-world applications. Code is available at https://github.com/BarSGuy/Window-Based-Distribution-Shift-Detection.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture:97086749422 & Meyer 1061
UID:eventx6a5a287eebd8710559
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20240313T121500
DTEND;TZID=Asia/Jerusalem:20240313T131500
DTSTAMP;TZID=Asia/Jerusalem:20240313T121500
SUMMARY: Theory Seminar  talk by Hilla Schefler (Technion)  Theory Seminar: A Unified Characterization Of Private Learning  at 2024-03-13 12:15:00
DESCRIPTION:Differential Privacy (DP) is a mathematical framewirk for ensuring the privacy of individuals in a dataset. Roughly speaking, it guarantees that privacy is protected in data analysis by ensuring that the output of an analysis does not reveal sensitive information about any specific individual, regardless of whether their data is included in the dataset or not.This talk presents a unified framework for characterizing both pure and approximate differentially private learnabiliity under the PAC model. The framework uses the language of graph theory: for a concept class H, we define the contradiction graph G of H. Its vertices are realizable datasets, and two datasets S, S′ are connected by an edge if they contradict each other (i.e., there is a point x that is labeled differently in S and S′). Our main finding is that the combinatorial structure of G is deeply related to learning H under DP. Learning H under pure DP is captured by the fractional clique number of G. Learning H under approximate DP is captured by the clique number of G. Consequently, we identify graph-theoretic dimensions that characterize DP learnability: the clique dimension and fractional clique dimension.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 201
UID:eventx6a5a287eebd9610560
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20240318T133000
DTEND;TZID=Asia/Jerusalem:20240318T143000
DTSTAMP;TZID=Asia/Jerusalem:20240318T133000
SUMMARY: TDC Seminar  talk by Raissa Nataf  TDC Seminar: Null Messages, Information and Coordination  at 2024-03-18 13:30:00
DESCRIPTION:I'll present our paper about the role that null messages play in synchronous systems with and without failures. Our work provides necessary and sufficient conditions on the structure of protocols for information transfer and coordination there. We start by introducing a new and more refined definition of null messages. A generalization of message chains that allow these null messages is provided and is shown to be necessary and sufficient for information transfer in reliable systems. Coping with crash failures requires a much richer structure, since not receiving a message may be the result of the sender’s failure. We introduce a class of communication patterns called resilient message blocks, which impose a stricter condition on protocols than the silent choirs of Goren and Moses (2020).\nOur paper has been published at DISC 2023.\n
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zisapel (ECE) 608
UID:eventx6a5a287eebda510563
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20240318T180000
DTEND;TZID=Asia/Jerusalem:20240318T200000
DTSTAMP;TZID=Asia/Jerusalem:20240318T180000
SUMMARY: CSpecial Event  talk by Yarden Zuckerman, SW Security Manager, Nvidia  Nvidia - Lecture And Pizza  at 2024-03-18 18:00:00
DESCRIPTION:Join us for a lecture and pizza!\nCyber Security Challenges In The Modern Era\nby Yarden Zuckerman, SW Security Manager, Nvidia\n\nMonday | March 18, 2024 | 18:00 p.m. | Piano Auditorium Taub Building\n\nRegister Here
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Auditorium 012, floor 0
UID:eventx6a5a287eebdb410562
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20240319T123000
DTEND;TZID=Asia/Jerusalem:20240319T143000
DTSTAMP;TZID=Asia/Jerusalem:20240319T123000
SUMMARY: CSpecial Event  Yahoo Is Coming To The Faculty On March 19 - Summer Internships And Research Opportunities  at 2024-03-19 12:30:00
DESCRIPTION:Yahoo holds a dedicated meeting with graduate students\n\nTuesday, March 19 at 12:30 p.m. at the CS Grads Club\n\nIn the program: an introduction to Yahoo's research in Israel and the summer internship program.\n\nRegister to the event here\n\nwaiting for you!!
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:CS Grads Club, Room 225
UID:eventx6a5a287eebdc410558
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20240320T113000
DTEND;TZID=Asia/Jerusalem:20240320T123000
DTSTAMP;TZID=Asia/Jerusalem:20240320T113000
SUMMARY: ceClub  talk by (George Washington University) Adam J Aviv   ceClub: From Dashboards To Labels: Helping Users Manage And Make Decision About Privacy  at 2024-03-20 11:30:00
DESCRIPTION:The surveillance economy, where tracking and collecting data on uses for the purpose of advertising and other actions, is central to much of the money-making enterprises of the modern technology ecosystem. Due to regulations and other forces, some of the largest companies, such as Google and Apple, have prioritized mechanisms for users to better manage and receive information about the kinds of data that is being collected about them. In this talk, I will explore how effective these mechanisms really are and ask the question, who are they really serving? I will present recent experiments we've performed on Google's data dashboards and their effectiveness, and also present ongoing work on Apple's app-based privacy nutrition labels, which describe apps functionality with relation to privacy. \n\nBio: Adam J. Aviv is an Associate Professor (with tenure) in the Department of Computer Science at the George Washington University and is the director of the GW-Usable Security (GWUSEC) Lab. He is currently on sabbatical during the 2023/2024 academic year and is a visiting scholar in the international school and the department of industrial engineering at Tel Aviv University. Dr. Aviv has published over 80 peer reviewed papers in areas related to computer security, privacy, and applied cryptography. Currently, his primary academic interests lie at the intersection of human computer interaction (HCI) and computer security and privacy, as well as research in network security and applied cryptography. He has made significant contributions in the space of mobile authentication, studying how users choose passwords and PINs for their mobile devices. Prior to GW, he was a assistant professor at the United States Naval Academy and a visiting assistant professor at Swarthmore College. Dr. Aviv received his B.S.E from Columbia University in the City of New York and his M.S.E. and Ph.D. from the University of Pennsylvania. He is a recipient of six NSF awards as a PI, including the prestigious NSF CAREER award.\n
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Meyer 861 &  Zoom Lecture:94673013539
UID:eventx6a5a287eebdd410564
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20240320T123000
DTEND;TZID=Asia/Jerusalem:20240320T143000
DTSTAMP;TZID=Asia/Jerusalem:20240320T123000
SUMMARY: CSpecial Event  Mobileye Spotlight Day  at 2024-03-20 12:30:00
DESCRIPTION:You are invited to the spotlight day of the Mobileye company at the Technion\nWednesday 20.3 | 12:30 | Taub Building
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub Building
UID:eventx6a5a287eebde810566
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20240320T131500
DTEND;TZID=Asia/Jerusalem:20240320T141500
DTSTAMP;TZID=Asia/Jerusalem:20240320T131500
SUMMARY: Theory Seminar  talk by Ilan Newman (University of Haifa)  Theory Seminar: Deterministic Online embedding of metrics  at 2024-03-20 13:15:00
DESCRIPTION:A finite metric space $(X,d)$ on a set of points $X$ is just the shortest path metric $d$ on a positively weighted graph $G=(X,E)$. In the online setting, the vertices of the input finite metric space $(X,d)$ are exposed one by one, together with their distances $d(*,*)$to the previously exposed vertices. The goal is to embed (map) $X$ into a given host metric space $(H,d_H)$ (finite or not) and so to distort the distances as little as possible (distortion is the worst case ratio of how the distance is expanded, assuming it is never contracted).I will start by a short survey on the main existing results of offline embedding into $ell_1, ell_2, ell_{infty}$ spaces. Then will present some results on online embedding: mainly lower bounds (on the best distortion) for small dimensional spaces, and some upper bounds for 2-dim Euclidean spaces.As an intriguing question: the "rigid" $K_5$ metric space is a metric space on $G=K_5$, in which each edge should be thought as a unit interval $[0,1]$. The points are the $5$ fixed vertices of $K_5$ (that are exposed first) in addition to $n$ points that are arbitrarily placed anywhere in the edges, and exposed one by one. It is easy to show that an offline embedding into the $2$-dim Euclidean results in a distortion $Omega(n)$. What can be achieved in the online case ?? It was "believed" that an exponential distortion could be proven. We show that the distortion is bounded by $O(n^2)$.The talk is based on joint work with Noam Licht and Yuri Rabinovich
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 201
UID:eventx6a5a287eebdf610568
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20240320T143000
DTEND;TZID=Asia/Jerusalem:20240320T153000
DTSTAMP;TZID=Asia/Jerusalem:20240320T143000
SUMMARY: pixel-club  talk by Elnatan Kadar (Graduate Seminar)  Pixel-Club: Explaining Classification by Image Decomposition  at 2024-03-20 14:30:00
DESCRIPTION:We propose a new way to explain and to visualize neural network classification through a decomposition-based explainable AI.\nInstead of providing an explanation heatmap, our method yields a decomposition of the image into class-agnostic and class-distinct parts, with respect to the data and chosen classifier. Following a fundamental signal processing paradigm of analysis and synthesis, the original image is the sum of the decomposed parts. We thus obtain a radically different way of explaining classification. The class-agnostic part ideally is composed of all image features which do not posses class information, where the class-distinct part is its complementary. This new visualization can be more helpful and informative in certain scenarios, especially when the attributes are dense, global and additive in nature, for instance, when colors or textures are essential for class distinction.\n\nM.Sc. student under the supervision of Pro. Guy Gilboa.\n
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Room 1061, EE Meyer Building 
UID:eventx6a5a287eebe0710565
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20240321T103000
DTEND;TZID=Asia/Jerusalem:20240321T113000
DTSTAMP;TZID=Asia/Jerusalem:20240321T103000
SUMMARY: MSC  talk by Amir Dachbash  The Altru-Egoistic Approach to Collaborative Caching  at 2024-03-21 10:30:00
DESCRIPTION:In this lecture we will explore collaborative caching algorithms in order to boost the effectiveness of caches in a distributed storage system. I'll introduce a scheme that partitions each node’s cache into two conceptual regions: an egoistic area whose goal is to contain the most valuable data for the node that owns the cache, and an altruistic area whose goal is to contain the most valuable data for the system as a whole. Each node’s division between these two regions is dynamically adjusted locally.We introduce a family of algorithms that analyze cross-nodes statistics to decide how much memory to allocate to each partition in each node. We study the behavior of these algorithms through simulations of both synthetic workloads as well as multiple real traces from several sources. These simulations demonstrate the improvement in cache hit ratio for the entire system compared to state-of-the-art\n
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture:93387090038
UID:eventx6a5a287eebe1710569
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20240326T113000
DTEND;TZID=Asia/Jerusalem:20240326T123000
DTSTAMP;TZID=Asia/Jerusalem:20240326T113000
SUMMARY: pixel-club  talk by Assaf Shocher, NVIDIA  Pixel-Club: Idempotent Generative Network  at 2024-03-26 11:30:00
DESCRIPTION:We propose a new approach for generative modeling based on training a neural network to be idempotent. An idempotent operator is one that can be applied sequentially without changing the result beyond the initial application, namely f(f(z))=f(z). The proposed model f is trained to map a source distribution (e.g, Gaussian noise) to a target distribution (e.g. realistic images) using the following objectives: (1) Instances from the target distribution should map to themselves, namely f(x)=x. We define the target manifold as the set of all instances that f maps to themselves. (2) Instances that form the source distribution should map onto the defined target manifold. This is achieved by optimizing the idempotence term, f(f(z))=f(z) which encourages the range of f(z) to be on the target manifold. Under ideal assumptions such a process provably converges to the target distribution. This strategy results in a model capable of generating an output in one step, maintaining a consistent latent space, while also allowing sequential applications for refinement. Additionally, we find that by processing inputs from both target and source distributions, the model adeptly projects corrupted or modified data back to the target manifold. This work is a first step towards a “global projector” that enables projecting any input into a target data distribution.Short bio: Assaf Shocher is a postdoctoral researcher at NVIDIA. Prior to that he was a postdoctoral fellow at UC Berkeley, working with Alyosha Efros, and a visiting researcher at Google. He received his PhD from the Weizmann Institute of Science, where he was advised by Michal Irani. He has a bachelor’s degrees in Physics and EE from Ben-Gurion University. His prizes and honors include the Rothschild postdoctoral fellowship, the Fulbright postdoctoral fellowship, John F. Kennedy award for outstanding Ph.D. at the Weizmann Institute, and the Blavatnik award for CS Ph.D. graduates.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Room 1061, EE Meyer Building 
UID:eventx6a5a287eebe2810572
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20240327T100000
DTEND;TZID=Asia/Jerusalem:20240327T143000
DTSTAMP;TZID=Asia/Jerusalem:20240327T100000
SUMMARY: CSpecial Event  Intel's tech experience is coming to campus!  at 2024-03-27 10:00:00
DESCRIPTION:Intel's tech experience is coming to campus!\nReady for the most innovative technological picnic you've ever seen?\nWe have loaded technological tools and our greatest minds to the track and we are on our way to you\n\nWednesday 27.3 | Mayer building, 3rd floor\n10:00 - AR | VR | HR Come and get to know us and our technologies\n12:30 - FPGA MicroPython\n\nThe number of places is limited, register at the link
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:EE Meyer Building 
UID:eventx6a5a287eebe3b10567
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20240327T113000
DTEND;TZID=Asia/Jerusalem:20240327T123000
DTSTAMP;TZID=Asia/Jerusalem:20240327T113000
SUMMARY: ceClub  talk by Eytan Modiano - Laboratory for Information and Decision Systems Massachusetts Institute of Technology
  CE-Club: Optimizing Information Freshness in Wireless Networks: From Theory to Implementation  at 2024-03-27 11:30:00
DESCRIPTION:Age of Information (AoI) is a recently proposed performance metric that captures the freshness of the information from the perspective of the application. AoI measures the time that elapsed from the moment that the most recently received packet was generated to the present time. In this talk, we explore the AoI optimization problem in wireless networks.We start by considering a wireless network with a number of nodes transmitting information to a base station and develop low-complexity transmission scheduling policies that result in near-optimal AoI performance. We then extend our results to wireless networks under general interference constraints, and develop joint routing and scheduling schemes for minimizing AoI. Finally, we discuss implementation of our transmission scheduling policies using software defined radios, and application to remote tracking of vehicles using UVAs.Sort bio: Eytan Modiano is The Richard C. Maclaurin Professor in the Department of Aeronautics and Astronautics and the Laboratory for Information and Decision Systems (LIDS) at MIT. Prior to Joining the faculty at MIT in 1999, he was a Naval Research Laboratory Fellow between 1987 and 1992, a National Research Council Post Doctoral Fellow during 1992-1993, and a member of the technical staff at MIT Lincoln Laboratory between 1993 and 1999. Eytan Modiano received his B.S. degree in Electrical Engineering and Computer Science from the University of Connecticut at Storrs in 1986 and his M.S. and PhD degrees, both in Electrical Engineering, from the University of Maryland, College Park, MD, in 1989 and 1992 respectively.His research is on modeling, analysis and design of communication networks and protocols. He received the Infocom Achievement Award (2020) for contributions to the analysis and design of cross-layer resource allocation algorithms for wireless, optical, and satellite networks. He is the co-recipient of the Infocom 2018 Best paper award, the MobiHoc 2018 best paper award, the MobiHoc 2016 best paper award, the Wiopt 2013 best paper award, and the Sigmetrics 2006 best paper award. He was the Editor-in-Chief for IEEE/ACM Transactions on Networking (2017-2020), and served as Associate Editor for IEEE Transactions on Information Theory and IEEE/ACM Transactions on Networking. He was the Technical Program co-chair for IEEE Wiopt 2006, IEEE Infocom 2007, ACM MobiHoc 2007, and DRCN 2015; and general co-chair of Wiopt 2021. He had served on the IEEE Fellows committee in 2014 and 2015, and is a Fellow of the IEEE and an Associate Fellow of the AIAA.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Meyer 861 
UID:eventx6a5a287eebe4c10573
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20240327T120000
DTEND;TZID=Asia/Jerusalem:20240327T133000
DTSTAMP;TZID=Asia/Jerusalem:20240327T120000
SUMMARY: CSpecial Event  talk by Prof. Daniel Jackson, MIT  Guest seminar - How to design innovative software?  at 2024-03-27 12:00:00
DESCRIPTION:I will explain how successful innovations in software can usually be traced to just one or two "concepts" that offer new scenarios that, with seemingly small shifts, radically change how an application is used. I give examples from apps such as Zoom, WhatsApp and Calendly. I explain how concepts, each with their own characteristic scenarios, can be composed to form apps. I explain how this idea can be used to make software more usable, modular and consistent.\n\nMore information at: http://essenceofsoftware.com\n\nProf. Jackson's bio: Daniel Jackson is professor of computer science at MIT, and associate director of the Computer Science and Artificial Intelligence Laboratory, MIT’s largest lab. His latest book, The Essence of Software, was published in November 2021, and offers a radical new approach to the design of software. Jackson is also a photographer. His book, Portraits of Resilience, which combined stories and photographic portraits of people who had experienced mental health challenges, was published by MIT Press in Dec 2017, and was featured on PBS's Newshour and NPR’s Here and Now.\n\nRegister at the link
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Auditorium 012, floor 0
UID:eventx6a5a287eebe6110574
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20240327T121500
DTEND;TZID=Asia/Jerusalem:20240327T131500
DTSTAMP;TZID=Asia/Jerusalem:20240327T121500
SUMMARY: Theory Seminar  talk by Tomer Koren (Tel-Aviv University)  Theory Seminar: Exploring the Generalization Ability of (Convex) Optimization Algorithms  at 2024-03-27 12:15:00
DESCRIPTION:In machine learning, there has been considerable interest over the past decade in understanding the ability of optimization algorithms to generalize—namely, to produce solutions (models) that extend well to unseen data—particularly in the context of overparameterized problems.  I will survey several recent theoretical studies that explore generalization in classical convex optimization, that reveal intriguing behavior of common optimization methods and shed some light on the concept of generalization in high dimension. 
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 201
UID:eventx6a5a287eebe7410575
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20240328T093000
DTEND;TZID=Asia/Jerusalem:20240328T153000
DTSTAMP;TZID=Asia/Jerusalem:20240328T093000
SUMMARY: CSpecial Event  Open day for March 28. Getting Started!  at 2024-03-28 09:30:00
DESCRIPTION:An open day for studies at the Technion will be held on Thursday, March 28, starting at 09:30\n\nFor details and registration, go to the registration and admission website at the link\n
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Technion
UID:eventx6a5a287eebe8310557
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20240401T123000
DTEND;TZID=Asia/Jerusalem:20240401T143000
DTSTAMP;TZID=Asia/Jerusalem:20240401T123000
SUMMARY: CSpecial Event  talk by StarkWare  StarkWare - Engineering the Future of Blockchain  at 2024-04-01 12:30:00
DESCRIPTION:Engineering the Future of Blockchain\n1.4.23 | 12:30-14:30 | The graduate students' lounge, Taub, 2nd floor\n\n12:30 - Mingling and Snacks\n13:00 - Oded Naor, Ph.D., Product Manager - Btockchain reinvented: a deep dive into StarkWare's game-changing sotutions\n13:30 - Noa Oved, MSC, Software Team Lead - Coding the future: Transforming algorithmic insights into real-wortd sotutions\n14:00- Q&amp;A\n\nPlese RSVP at the link
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:The graduate students' lounge, room 225
UID:eventx6a5a287eebe9110576
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20240402T113000
DTEND;TZID=Asia/Jerusalem:20240402T123000
DTSTAMP;TZID=Asia/Jerusalem:20240402T113000
SUMMARY: pixel-club  talk by Anton Agafonov (Graduate Seminar)  Pixel-Club: STMPL: Human Soft-Tissue Simulation  at 2024-04-02 11:30:00
DESCRIPTION:In various applications, such as virtual reality and gaming, simulating the deformation of soft tissues in the human body during interactions with external objects is essential. Traditionally, Finite Element Methods (FEM) have been employed for this purpose, but they tend to be slow and resource-intensive. In this paper, we propose a unified representation of human body shape and soft tissue with a data-driven simulator of non-rigid deformations. This approach enables rapid simulation of realistic interactions. Our method builds upon the SMPL model, which generates human body shapes considering rigid transformations. We extend SMPL by incorporating a soft tissue layer and an intuitive representation of external forces applied to the body during object interactions. Specifically, we mapped the 3D body shape and soft tissue and applied external forces to 2D UV maps. Leveraging a UNET architecture designed for 2D data, our approach achieves high-accuracy inference in real time. Our experiment shows that our method achieves plausible deformation of the soft tissue layer, even for unseen scenarios.\nM.Sc. student under the supervision of Prof. Lihi Zelnik-Manor.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Room 608, Zisapel Building
UID:eventx6a5a287eebea310578
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20240402T143000
DTEND;TZID=Asia/Jerusalem:20240402T153000
DTSTAMP;TZID=Asia/Jerusalem:20240402T143000
SUMMARY: colloq  talk by Hadar Averbuch-Elor, Tel-Aviv University  Marrying Vision and Language: A Mutually Beneficial Relationship?  at 2024-04-02 14:30:00
DESCRIPTION:Foundation models that connect vision and language have recently shown great promise for a wide array of tasks such as text-to-image generation. Significant attention has been devoted towards utilizing the visual representations learned from these powerful vision and language models. In this talk, I will present an ongoing line of research that focuses on the other direction, aiming at understanding what knowledge language models acquire through exposure to images during pretraining. We first consider in-distribution text and demonstrate how multimodally trained text encoders, such as that of CLIP, outperform models trained in a unimodal vacuum, such as BERT, over tasks that require implicit visual reasoning. Expanding to out-of-distribution text, we address a phenomenon known as sound symbolism, which studies non-trivial correlations between particular sounds and meanings across languages, and demonstrate the presence of this phenomenon in vision and language models such as CLIP and Stable Diffusion. Our work provides new angles for understanding what is learned by these vision and language foundation models, offering principled guidelines for designing models for tasks involving visual reasoning.Short Bio: Hadar Averbuch-Elor is an Assistant Professor at the School of Electrical Engineering in Tel Aviv University. Before that, Hadar was a postdoctoral researcher at Cornell-Tech, working with Noah Snavely. She completed her PhD in Electrical Engineering at Tel-Aviv University, where she was advised by Daniel Cohen-Or. Hadar is a recipient of multiple awards including the Zuckerman Postdoctoral Scholar Fellowship, the Schmidt Postdoctoral Award for Women in Mathematical and Computing Sciences, and the Alon Scholarship. She was also selected as a Rising Star in EECS by UC Berkeley. Hadar's research interests lie in the intersection of computer graphics and computer vision, particularly in combining pixels with more structured modalities, such as natural language and 3D geometry.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 337
UID:eventx6a5a287eebeb510570
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20240403T113000
DTEND;TZID=Asia/Jerusalem:20240403T123000
DTSTAMP;TZID=Asia/Jerusalem:20240403T113000
SUMMARY: ceClub  talk by Ben Nassi, Technion  CE-Club:Here Comes the GenAI Malware, Worm, and APT  at 2024-04-03 11:30:00
DESCRIPTION:In the past year, numerous companies have incorporated Generative AI (GenAI) capabilities into new and existing applications, forming interconnected Generative AI (GenAI) ecosystems consisting of applications powered by GenAI services.While ongoing research highlighted risks associated with the GenAI layer of agents (e.g., dialog poisoning, membership inference, prompt leaking, jailbreaking), a critical question emerges: Can attackers develop malware to exploit the GenAI component of an application and launch cyber-attacks on the GenAI-powered application or the entire GenAI ecosystem?In this talk, we discuss the new evolving attack landscape of input prompts being used to conduct malicious activities in the context of an application.We show how attackers can craft adversarial self-replicating prompts, which when sent to GenAI-powered applications can form as (1) malware that launches a DoS attack against a GenAI-powered application by causing it to enter an infinite loop, (2) a worm that extracts a user's sensitive information from GenAI-powered email assistants and compromises new GenAI-powered applications, and (3) APT (advanced persistent threat) that uses the advanced AI capabilities of the GenAI model to identify the assets in the context, determine the possible malicious activities to conduct, execute one of them, and cover its tracks.The findings in our study are demonstrated against three different applications powered by three different GenAI models (Gemini Pro, ChatGPT 4.0, and LLaVA).\nAt the end of the talk, we discuss the upcoming changes in application security that are about to appear in the next few years due to the increased integration of GenAI capabilities into existing and new applications.\n
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Meyer 861 
UID:eventx6a5a287eebeca10582
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20240403T121500
DTEND;TZID=Asia/Jerusalem:20240403T131500
DTSTAMP;TZID=Asia/Jerusalem:20240403T121500
SUMMARY: Theory Seminar  talk by Or Meir (Haifa University)   Theory Seminar: Toward Better Depth Lower Bounds: A KRW-like theorem for Strong Composition  at 2024-04-03 12:15:00
DESCRIPTION:One of the major open problems in complexity theory is proving super-logarithmic lower bounds on the depth of circuits. Karchmer, Raz, and Wigderson (Computational Complexity 5(3/4), 1995) suggested approaching this problem by proving that depth complexity of a composition of two functions is roughly the sum of their individual depth complexities. They showed that the validity of this conjecture would imply the desired lower bounds.The intuition that underlies the KRW conjecture is that composition should behave like a “direct-sum problem”, in a certain sense, and therefore the depth complexity of the composition should be the sum of the individual depth complexities. Nevertheless, there are two obstacles toward turning this intuition into a proof: first, we do not know how to prove that the composition must behave like a direct-sum problem; second, we do not know how to prove that the complexity of the latter direct-sum problem is indeed the sum of the individual complexities.In this talk, we will describe a new work that tackles the second obstacle. We will consider a notion of “strong composition” that is forced to behave like a direct-sum problem, and see that a variant of the KRW conjecture holds for this notion. This result demonstrates that the first obstacle above is the crucial barrier toward resolving the KRW conjecture. Along the way, we will discuss some general techniques that might be of independent interest.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 201
UID:eventx6a5a287eebedf10579
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20240403T123000
DTEND;TZID=Asia/Jerusalem:20240403T143000
DTSTAMP;TZID=Asia/Jerusalem:20240403T123000
SUMMARY: CSpecial Event  CS RESEARCH DAY 2024  at 2024-04-03 12:30:00
DESCRIPTION:The 12th CS Research Day for graduate studies will be held on Wednesday, April 03, 2024 between 12:30-14:30, at the lobby of the CS Taub Building.\n\nResearch Day events are opportunity for our graduate students to expose their researches using posters and presentations to CS faculty and all degrees students, Technion distinguished representatives and to high-ranking delegates from the hi-tech leading industry companies in Israel and abroad. \n\nThe participating researches will be on various topics: Cryptology and Cyber, Data Centers and Clouds, Graphics, Intelligent Systems and Scientific Computation, Machine Learning and Information Retrieval, Systems and Applications, Testing and Verification, Theory of Computer Science.\n\nThe presenting researches
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:CS Taub Lobby
UID:eventx6a5a287eebef310542
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20240404T103000
DTEND;TZID=Asia/Jerusalem:20240404T113000
DTSTAMP;TZID=Asia/Jerusalem:20240404T103000
SUMMARY: colloq  talk by Gal Chechik,  Bar-Ilan University and NVIDIA  Learning with visual foundation models for Gen AI  at 2024-04-04 10:30:00
DESCRIPTION:Between training and inference, lies a growing class of AI problems that involve fast optimization of a pre-trained model for a specific inference task. These are not pure “feed-forward” inference problems applied to a pre-trained model, because they involve some non-trivial inference-time optimization beyond what the model was trained for; neither are they training problems, because they focus on a specific input. These compute-heavy inference workflows raise new challenges in machine learning and open opportunities for new types of user experiences and use cases.In this talk, I describe two main flavors of the new workflows in the context of text-to-image generative models: few-shot fine-tuning and inference-time optimization. I'll cover personalization of vision-language models using textual-inversion techniques, and techniques for model inversion, prompt-to-image alignment and consistent generation.  I will also discuss the generation of rare classes, and future directions.Short Bio: Gal Chechik is a Professor of computer science at Bar-Ilan University and a senior director of AI research at NVIDIA. His current research focuses on learning for reasoning and perception. In 2018, Gal joined NVIDIA to found and head nvidia's research in Israel. Prior to that, Gal was a staff research scientist at Google Brain and Google research developing large-scale algorithms for machine perception, used by millions daily. Gal earned his PhD in 2004 from the Hebrew University, and completed his postdoctoral training at Stanford CS department.  Gal authored ~130 refereed publications, ~50 patents, including publications in Nature Biotechnology, Cell and PNAS. His work won awards at ICML and NeurIPS.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 337
UID:eventx6a5a287eebf0510571
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20240407T100000
DTEND;TZID=Asia/Jerusalem:20240407T110000
DTSTAMP;TZID=Asia/Jerusalem:20240407T100000
SUMMARY: MSC  talk by Zachary Elisha Bamberger  DEPTH: Discourse Education through Pre-Training Heirarchically  at 2024-04-07 10:00:00
DESCRIPTION:Language Models (LMs) excel in many tasks, but understanding discourse – how sentences connect to form coherent text – remains a challenge. This is especially true for smaller models aiming to match the abilities of their larger counterparts in handling long and complex inputs. To address this, we introduce DEPTH, a new encoder-decoder model designed to foster robust discourse-level representations during the pre-training phase. DEPTH uniquely combines hierarchical sentence representations and the “Sentence Un-shuffling” task with traditional span-corruption objective of encoder-decoder LMs. While span-corruption helps the model learn word-level dependencies, ”Sentence Un- shuffling” forces it to restore the natural order of scrambled sentences, teaching it about the logical flow of language. The encoder-decoder architecture allows DEPTH to consider words both before and after a given token, offering a more nuanced contextual understanding than decoder-only models like GPT. This is crucial for tasks that depend on how sentences interrelate.We built a pre-training and fine-tuning framework for encoder-decoder models that facilitated our experiments with both T5 and DEPTH, and that seamlessly integrates with open-source tools for distributed training in the HuggingFace ecosystem. DEPTH’s training is remarkably computationally efficient, as it learns meaningful semantic- and discourse-level representations at a faster rate than it’s T5 counterpart. Notably, already in early stages of the pre-training phase, we find that DEPTH outperforms T5 by reaching a lower span-corruption loss. This occurs despite the fact that T5 is trained solely with this objective, and DEPTH is tasked with the additional sentence un-shuffling objective. Evaluations on GLUE and DiscoEval benchmarks demonstrate DEPTH’s ability to quickly learn downstream tasks spanning understanding of syntax (CoLA), sentinment analysis (SST2), sentence positioning (SP), discourse coherence (DC), and natural language inference (MNLI).
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Room 601
UID:eventx6a5a287eebf1610577
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20240408T133000
DTEND;TZID=Asia/Jerusalem:20240408T143000
DTSTAMP;TZID=Asia/Jerusalem:20240408T133000
SUMMARY: TDC Seminar  talk by Michal Dory, Haifa University  TDC Seminar: Fault-Tolerant Labeling and Compact Routing Schemes  at 2024-04-08 13:30:00
DESCRIPTION:Assume that you have a huge graph, where edges and vertices may fail, and you want to answer questions about this graph quickly, without inspecting the whole graph. A fault-tolerant labeling scheme allows us to do just that. It is a distributed data structure, in which each vertex and edge is assigned a short label, such that given the labels of a pair of vertices s,t, and a set of failures F, you can answer questions about s and t in G\F. For example, determine if s and t are connected in G\F, just by looking at their labels.I will discuss fault-tolerant connectivity labeling schemes for general graphs. I will also show applications for fault-tolerant routing. Here the goal is to provide each vertex a small routing table, such that given these tables we can efficiently route messages in the network, even in the presence of failures.Based on a joint work with Merav Parter.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zisapel (ECE) 608
UID:eventx6a5a287eebf2710580
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20240409T110000
DTEND;TZID=Asia/Jerusalem:20240409T120000
DTSTAMP;TZID=Asia/Jerusalem:20240409T110000
SUMMARY: MSC  talk by Nadav Rubinstein  PAT-CEP : Statistical Guarantee for ML-based Complex Event Processing  at 2024-04-09 11:00:00
DESCRIPTION:In the landscape of data stream processing, the challenges posed by online problems are of paramount importance. This paper introduces a pioneering method tailored to address these challenges within the context of data streams. While our initial focus centered on Complex Event Processing (CEP) problems, it is essential to underscore the versatility of our approach, as it is equally applicable to a diverse range of online problems. To the best of our knowledge, there exists no comprehensive method in the existing literature that comprehensively tackles the intricacies of data streams in online problem-solving. This paper elucidates our innovative approach and its potential to redefine problem-solving methodologies across various online domains.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture:94644097135
UID:eventx6a5a287eebf3510583
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20240409T113000
DTEND;TZID=Asia/Jerusalem:20240409T123000
DTSTAMP;TZID=Asia/Jerusalem:20240409T113000
SUMMARY: pixel-club  talk by Tom Bekor - The Andrew and Erna Viterbi Faculty of Electrical & Computer Engineering
  Pixel Club: FreeAugment: Data Augmentation Search Across All Degrees of Freedom  at 2024-04-09 11:30:00
DESCRIPTION:Graduate Seminar\n\nData augmentation has become an integral part of deep learning, as it is known to improve the generalization capabilities of neural networks.Since the most effective set of image transformations differs between tasks and domains, automatic data augmentation search aims to alleviate the extreme burden of manually finding the optimal image transformations. However, current methods are not able to jointly optimize all degrees of freedom: (1) the number of transformations to be applied, their (2) types, (3) order, and (4) magnitudes. Many existing methods risk picking the same transformation more than once, limit the search to two transformations only, or search for the number of transformations exhaustively or iteratively in a myopic manner.Our approach, FreeAugment, is the first to achieve global optimization of all four degrees of freedom simultaneously, using a fully differentiable method. It efficiently learns the number of transformations and a probability distribution over their permutations, inherently refraining from redundant repetition while sampling.Our experiments demonstrate that this joint learning of all degrees of freedom significantly improves performance, achieving state-of-the-art results on various natural image benchmarks and across other domains.M.Sc. student under the supervision of Prof. Lihi Zelnik-Manor.\n
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Room 1061, EE Meyer Building 
UID:eventx6a5a287eebf4410584
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20240410T113000
DTEND;TZID=Asia/Jerusalem:20240410T123000
DTSTAMP;TZID=Asia/Jerusalem:20240410T113000
SUMMARY: ceClub  talk by Yara Mulla ( NVidia - IBC Barracuda)
  CE-Club: Buffer Isolation With Imperfect Congestion Control Classification  at 2024-04-10 11:30:00
DESCRIPTION:Due to their increasing aggressiveness, recent congestion control algorithms (CCAs) can quickly starve standard TCP flows in their shared router queues. Existing solutions based on fair queueing are not scalable enough, and those based on admission control do not fit all CCAs. Independently, building on the popularization of machine learning, recent papers have designed several CCA classifiers.In this seminar, we introduce a buffer isolation mode where incoming flows first undergo CCA classification, and then are mapped to distinct CCA-based queues based on their classified CCA. We provide a fundamental analysis for such an isolation mode, and present a simple and an advanced aggressiveness model for its performance. Then, in evaluations, we show how this isolation mode clearly outperforms the existing sharing mode, as well as how our advanced model can accurately represent its performance. We further show how using the advanced model, we can maximize the performance by optimizing the queue service rates, surprisingly obtaining fair CCA shares with imperfect classification.Yara is an MSc student supervised by Prof. Isaac Keslassy.\n\nThe seminar will be given in Hebrew.\n
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture: 83464479226\n
UID:eventx6a5a287eebf5410585
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20240415T100000
DTEND;TZID=Asia/Jerusalem:20240415T140000
DTSTAMP;TZID=Asia/Jerusalem:20240415T100000
SUMMARY: CSpecial Event  A VIP Visit At The Technion For Children Of Graduates  at 2024-04-15 10:00:00
DESCRIPTION:Graduates of the faculty are invited with their children to a day of lectures, workshops and information regarding registration and admission - for children; and a nostalgic tour (for parents)\nMonday, April 15, starting at 10:00 at the Technion\n\nIn the program:\n1. The wonders of infinity - the Faculty of Mathematics\n2. What is the Internet of Things? - Faculty of Computer Science\n3. Games and Virtual Reality - Faculty of Architecture and Urban Planning\n\nRegister at the link (the number of places is limited)\n\nThe activity is intended for ages 16 and older.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Technion, Ulman Building
UID:eventx6a5a287eebf6510581
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20240415T140000
DTEND;TZID=Asia/Jerusalem:20240415T150000
DTSTAMP;TZID=Asia/Jerusalem:20240415T140000
SUMMARY: MSC  talk by David Valensi  Deep Learning and Tree Search-Based Policy Optimization under Stochastic Execution Delay  at 2024-04-15 14:00:00
DESCRIPTION:The standard formulation of Markov decision processes (MDPs) assumes that the agent's decisions are executed immediately. However, in numerous realistic applications such as robotics or healthcare, actions are performed with a delay whose value can even be stochastic. In this work, we introduce stochastic delayed execution MDPs, a new formalism addressing random delays without resorting to state augmentation. We show that given observed delay values, it is sufficient to perform a policy search in the class of Markov policies in order to reach optimal performance, thus extending the deterministic fixed delay case. Armed with this insight, we devise DEZ, a model-based algorithm that optimizes over the class of Markov policies. DEZ leverages Monte-Carlo tree search similar to its non-delayed variant EfficientZero to accurately infer future states from the action queue. Thus, it handles delayed execution while preserving the sample efficiency of EfficientZero. Through a series of experiments on the Atari suite, we demonstrate that although the previous baseline outperforms the naive method in scenarios with constant delay, it underperforms in the face of stochastic delays. In contrast, our approach significantly outperforms the baselines, for both constant and stochastic delays.This work was accepted to the conference ICLR 2024.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zisapel 413 &  Zoom Lecture: 98471388934
UID:eventx6a5a287eebf7810586
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20240416T143000
DTEND;TZID=Asia/Jerusalem:20240416T153000
DTSTAMP;TZID=Asia/Jerusalem:20240416T143000
SUMMARY: pixel-club  talk by Elias Nehme - The Andrew and Erna Viterbi Faculty of Electrical & Computer Engineering
  Pixel Club: Deep Computational Imaging: Optimal sensing, reconstruction, and uncertainty quantification  at 2024-04-16 14:30:00
DESCRIPTION:In biological imaging, fast acquisition of depth information is crucial e.g. for accurate 3D tracking of sub-cellular elements and for 3D super-resolution. In the first part of this talk, we present a series of works enhancing the success of snapshot depth sensing in the revolutionary field of single-molecule localization microscopy (Nobel Prize in Chemistry 2014). Specifically, we present an approach for jointly designing the “optics” of the microscope and the 3D reconstruction algorithm, by using deep learning. Our approach is demonstrated experimentally with super-resolution reconstructions of mitochondria and volumetric imaging and tracking of fluorescently labelled telomeres in live cancer cells. In the second part of this talk, we focus on the challenging task of visualizing and quantifying reconstruction uncertainty in computational imaging systems. Specifically, we present two techniques for visualizing prediction uncertainty in imaging inverse problems. Our methods are at least as accurate, and orders of magnitude faster than baselines that rely on state-of-the-art posterior samplers. We demonstrate the practical benefit of our methods for severely ill-posed inverse problems, including transferring the images of a biological sample imaged with one fluorescent dye to appear as if they were imaged with another.Elias Nehme is a PhD candidate at the ECE department, jointly supervised by prof. Tomer Michaeli and prof. Yoav Shechtman. Prior to that he received his bachelor’s in biomedical engineering also from the Technion. His awards and honors include the Lev-Margulis memorial prize in microscopy, the best poster award in quantitative bioimaging, the Jacobs-Qualcomm fellowship in 3D imaging and reconstruction, and several VATAT prizes in data science.Ph.D. Under the supervision of Prof. Tomer Michaeli and  Prof. Yoav Shechtman.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Room 1061, EE Meyer Building  &  Zoom Lecture:5049603009
UID:eventx6a5a287eebf8d10587
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20240421T160000
DTEND;TZID=Asia/Jerusalem:20240421T170000
DTSTAMP;TZID=Asia/Jerusalem:20240421T160000
SUMMARY: MSC  talk by Tom Azoulay  Blockchain Loans: Mechanisms, Platforms And Challenges  at 2024-04-21 16:00:00
DESCRIPTION:Distributed lending protocols are major financial mechanisms that match borrowers and lenders without the need for intermediaries. Smart contracts deployed on blockchains ensure the security of such loans. Borrowers mitigate the risk of default by providing collateral in the form of cryptocurrencies or by limiting the loan duration. \nIn the first part of the talk, we describe various loan types, major platforms, and financial models that allow them. In the second part of the talk, we refer to a major challenge in collateral-based loans. The high volatility of cryptocurrencies implies a serious barrier of entry with a common practice that collateral values equal multiple times the loan's value. This requirement prevents many loan candidates from obtaining them.  We study a new approach that lowers capital requirements and augments inclusiveness with a low added risk. Risk estimation is crucial to guarantee the ongoing operation of the lending protocols, allowing them to be selective in providing loans and setting interest rates. We describe a risk estimation of high accuracy for borrowers based on their past activity. We evaluate our methodology on data collected directly from the Ethereum Network and show it achieves high gains and estimation accuracy.\nA part of this work appeared at the IEEE International Conference on Blockchain and Cryptocurrency (ICBC), 2023\n
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture
UID:eventx6a5a287eebfa210588
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20240505T150000
DTEND;TZID=Asia/Jerusalem:20240505T160000
DTSTAMP;TZID=Asia/Jerusalem:20240505T150000
SUMMARY: MSC  talk by Mousa Arraf  CIQA : A Coding Inspired Question Answering Model  at 2024-05-05 15:00:00
DESCRIPTION:Methods in question-answering (QA) that transform texts detailing processes into an intermediate code representation, subsequently executed to generate a response to the presented question, have demonstrated promising results in analyzing scientific texts that describe intricate processes.\nThe limitations of these existing text-to-code models are evident when attempting to solve QA problems that require knowledge beyond what is presented in the input text. We propose a novel domain-agnostic model to address the problem by leveraging domain-specific and open-source code libraries. \nWe introduce an innovative QA text-to-code algorithm that learns to represent and utilize external APIs from code repositories, such as GitHub, within the intermediate code representation. The generated code is then executed to answer a question about a text.\nWe present three QA datasets, focusing on scientific problems in the domains of chemistry, astronomy, and biology, for the benefit of the community.\nOur study demonstrates that our proposed method is a competitive alternative to current state-of-the-art (SOTA) QA text-to-code models and generic SOTA QA models.\n
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture:91070061962
UID:eventx6a5a287eebfb610591
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20240515T153000
DTEND;TZID=Asia/Jerusalem:20240515T163000
DTSTAMP;TZID=Asia/Jerusalem:20240515T153000
SUMMARY: MSC  talk by Omer Yizhaq  Jumping Automata over Infinite Words  at 2024-05-15 15:30:00
DESCRIPTION:We introduce and study jumping automata over infinite words, a fascinating twist on traditional finite automata. These machines read their input in a non-consecutive manner, defying conventional word order. We explore three distinct semantics: one ensuring every letter is accounted for, another permitting word permutation within fixed windows, and a third allowing permutation within windows of an existentially-quantified bound. Our work covers expressiveness, closure properties, algorithmic characteristics of these models, and more.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom lecture: 95929737670, password: OmerMSC and room 601
UID:eventx6a5a287eebfc910590
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20240516T140000
DTEND;TZID=Asia/Jerusalem:20240516T150000
DTSTAMP;TZID=Asia/Jerusalem:20240516T140000
SUMMARY: PHD  talk by Shaked Brody  Structural Language Models for Code  at 2024-05-16 14:00:00
DESCRIPTION:In the past few years, software has been at the heart of many applications ranging from appliances to virtual services. Helping software developers to write better code is a crucial task. In parallel, recent developments in machine learning and deep learning in particular have shown great promise in many fields, and code-related tasks in particular. The main challenge is how to represent code in a way that can be used by deep learning models effectively. While code can be treated as a sequence of tokens, it also can be represented using its underlying Abstract Syntax Tree (AST) that contains rich structural information. In this thesis, we investigate the use of the structural nature of code for code-related tasks.We start by introducing the edit completion task, a new task that requires predicting the next edit operation in a code snippet, given a sequence of contextual edits. This task may be helpful for software developers, as a substantial part of the time spent on writing code is dedicated to editing existing code. We investigate different approaches for this task and show that harnessing the structural nature of code can be beneficial for this task. Then, we generalize different structural approaches for addressing code-related tasks and show that the combined approach can be beneficial for different tasks such as edit completion and code summarization. The use of Graph Neural Networks (GNNs) is common for structural code representation. We reveal that the commonly used Graph Attention Network (GAT) model has a limitation in capturing complex relationships in graphs due to its attention mechanism. We provide an analysis of the expressive power of GAT and introduce a simple fix to it, GATv2, that has proven to be more powerful. Our experiments show that GATv2 gains better performance in different tasks in is more robust to noisy graphs. The Transformer architecture is widely used in the natural language processing (NLP) field as a model for sequence processing. However, this architecture can be considered as a special case of GNNs. Layer Normalization is a key component in the Transformer architecture. In the last part of this thesis, we investigate the expressivity role of Layer Normalization in the Transformers' attention. We provide a novel geometric interpretation of Layer Normalization and show its importance to the attention mechanism that follows it in the Transformer architecture. Our work provides both practical and theoretical contributions to the field of using deep neural models on code.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture:92691399579
UID:eventx6a5a287eebfdb10592
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20240522T113000
DTEND;TZID=Asia/Jerusalem:20240522T123000
DTSTAMP;TZID=Asia/Jerusalem:20240522T113000
SUMMARY: ceClub  talk by Yekaterina Podiatchev   CE-Club: Survivable Payment Channel Networks  at 2024-05-22 11:30:00
DESCRIPTION:Payment Channel Networks (PCNs) are a leading method to scale the transaction throughput in cryptocurrencies. Two participants can use a bidirectional payment channel for making multiple mutual payments without committing them to the blockchain. Opening a payment channel is a slow operation that involves an on-chain transaction locking a certain amount of funds. These aspects limit the number of channels that can be opened or maintained. Users may route payments through a multi-hop path and thus avoid opening and maintaining a channel for each new destination. Unlike regular networks, in PCN capacity depends on the usage patterns and, moreover, channels may become unidirectional. Since payments often fail due to channel depletion, a protection scheme to overcome failures is of interest. We define the stopping time of a payment channel as the time at which the channel becomes depleted. We analyze the mean stopping time of a channel as well as that of a network with a set of channels and examine the stopping time of channels in particular topologies. We then propose a scheme for optimizing the capacity distribution among the channels in order to increase the minimal stopping time in the network. We conduct experiments and demonstrate the accuracy of our model and the efficiency of the proposed optimization scheme.Yekaterina is an M.Sc. student supervised by Prof. Ori Rottenstreich and Prof. Ariel Orda.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Meyer 861 
UID:eventx6a5a287eebff310595
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20240522T130000
DTEND;TZID=Asia/Jerusalem:20240522T150000
DTSTAMP;TZID=Asia/Jerusalem:20240522T130000
SUMMARY: CSpecial Event  The Sky Is The Limit - An Open Day For Advanced Degree Studies At The Faculty  at 2024-05-22 13:00:00
DESCRIPTION:A meeting of those interested in graduate studies at the Taub Faculty of Computer Science at the Technion will be held on Wednesday, May 22, 2024, 1:00 p.m., room 337\n\nOutstanding bachelor's graduates from all universities!\nThis is your opportunity to participate and be impressed by the Faculty of Computer Science at the Technion, Meet faculty members and graduate students and hear fascinating lectures.\n\nEvent schedule:\n\n13:00-13:30 Opening remarks:\nThe words of the dean of the faculty, Prof. Danny Raz\nThe words of Vice Dean for Graduate Studies, Prof. Gill Barequet\n\n13:30-14:10 Lectures:\nDr. Jonathan Yaniv, graduate of the faculty and Head of Data Science &amp; ML Engineering at Yotpo: "From studies to career"\nMr. Dean Zadok (doctoral student): Life in the Faculty of Computer Science\n\n14:10 Questions and answers\n\nTo participate, please register in advance at the link\n\nFor questions and details, please contact the graduate studies coordinator, Anna Kleiner
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 337
UID:eventx6a5a287eec00910589
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20240529T113000
DTEND;TZID=Asia/Jerusalem:20240529T123000
DTSTAMP;TZID=Asia/Jerusalem:20240529T113000
SUMMARY: MSC  talk by Julian Mour  Robustness Verification of Multi-Label Neural Network Classifiers  at 2024-05-29 11:30:00
DESCRIPTION:Multi-label neural networks are important in various tasks, including safety-critical tasks. Several works show that these networks are susceptible to adversarial attacks, which can remove a target label from the predicted label list or add a target label to this list. However, no verifier can deterministically determine the list of labels for which a multi-label neural network is locally robust. The main challenge is that the complexity of the analysis increases by a factor exponential in the number of predicted classes and the total number of classes.We propose MuLLoC, a sound and complete robustness verifier for multi-label image classifiers that determines the robust labels in a given neighborhood of inputs. \nTo scale the analysis, MuLLoC relies on fast, optimistic queries to the network or to a constraint solver. Its queries include sampling and pair-wise relation analysis via numerical optimization and mixed-integer linear programming (MILP). For the remaining unclassified labels, MuLLoC performs an exact analysis by a novel MILP encoding for multi-label classifiers.We evaluate MuLLoC on three multi-label image datasets and several convolutional networks. Our results show that MuLLoC classifies all labels as robust or not within 17 minutes on average and that sampling and pair-wise relation analysis classify 94.62% of the labels.\n\n
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Meyer 861 &  Zoom Lecture: 91074303117
UID:eventx6a5a287eec01e10594
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20240529T113000
DTEND;TZID=Asia/Jerusalem:20240529T123000
DTSTAMP;TZID=Asia/Jerusalem:20240529T113000
SUMMARY: ceClub  talk by Julian Moure  CE-Club: Robustness Verification of Multi-Label Neural Network Classifiers  at 2024-05-29 11:30:00
DESCRIPTION:Multi-label neural networks are important in various tasks, including safety-critical tasks. Several works show that these networks are susceptible to adversarial attacks, which can remove a target label from the predicted label list or add a target label to this list. However, no verifier can deterministically determine the list of labels for which a multi-label neural network is locally robust. The main challenge is that the complexity of the analysis increases by a factor exponential in the number of predicted classes and the total number of classes. We propose MuLLoC, a sound and complete robustness verifier for multi-label image classifiers that determines the robust labels in a given neighborhood of inputs. To scale the analysis, MuLLoC relies on fast, optimistic queries to the network or to a constraint solver. Its queries include sampling and pair-wise relation analysis via numerical optimization and mixed-integer linear programming (MILP). For the remaining unclassified labels, MuLLoC performs an exact analysis by a novel MILP encoding for multi-label classifiers. We evaluate MuLLoC on three multi-label image datasets and several convolutional networks. Our results show that MuLLoC classifies all labels as robust or not within 17 minutes on average and that sampling and pair-wise relation analysis classify 94.62% of the labels.Julian is an MSc student supervised by Dr. Dana Drechsler-Cohen.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture: 91074303117 & Meyer 861 
UID:eventx6a5a287eec03210599
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20240529T124500
DTEND;TZID=Asia/Jerusalem:20240529T134500
DTSTAMP;TZID=Asia/Jerusalem:20240529T124500
SUMMARY: Theory Seminar  talk by Theory group members and interested people  Theory Seminar  at 2024-05-29 12:45:00
DESCRIPTION:This week we plan to get acclimated to our seminar's new location (outside of Taub, due to construction work), in a more social atmosphere. This includes introductions of current and prospective members of the group,* and talk about plans for the semester, including theory courses scheduled. Come meet established theorists and folks interested in theory research!\n\n* if you know students doing or interested in theory who are not (or might not be) registered to this mailing list, please let them know about it!\n
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Amado 619 
UID:eventx6a5a287eec04510600
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20240531T100000
DTEND;TZID=Asia/Jerusalem:20240531T130000
DTSTAMP;TZID=Asia/Jerusalem:20240531T100000
SUMMARY: CSpecial Event  Open day for undergraduate studies at the Technion May 31 in Sarona, Tel Aviv  at 2024-05-31 10:00:00
DESCRIPTION:An open day for undergraduate studies at the Technion will be held on May 31 from 10:00 am to 1:00 pm in Sarona, Tel Aviv.\n\nAt the event you can meet the representatives of the faculty, consult, get an impression and get answers about the studies.\n\nRegister for the open day at the link\n
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Sarona, Tel Aviv
UID:eventx6a5a287eec05510598
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20240602T133000
DTEND;TZID=Asia/Jerusalem:20240602T143000
DTSTAMP;TZID=Asia/Jerusalem:20240602T133000
SUMMARY: PHD  talk by Gal Sela  Efficient, Correct and Durable Concurrent Algorithms  at 2024-06-02 13:30:00
DESCRIPTION:In the last two decades, a main way to improve performance of computer processors has been producing processors with multiple cores, on which tasks may execute concurrently. Software is required to keep up with the hardware advances and supply concurrent algorithms for programs that run on multiple cores. This talk will focus on a fundamental building block of concurrent algorithms: concurrent data structures, designed with improved efficiency while also satisfying strong correctness and progress guarantees and robustness to failures.\n
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture:97432970957
UID:eventx6a5a287eec06710596
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20240604T183000
DTEND;TZID=Asia/Jerusalem:20240604T203000
DTSTAMP;TZID=Asia/Jerusalem:20240604T183000
SUMMARY: CSpecial Event  The CTF Group Of The Faculty - Technipwn - Invites You To The Opening Lecture Of The Semester 4.6.2024  at 2024-06-04 18:30:00
DESCRIPTION:The CTF group of the faculty - Technipwn invites you to the semester opening lecture - a year of offensive security: attacks on OpenAI, Atlassian, Apple and more! \n\nSpeaker>> Ron Massas, Vulnerability Researcher at Imperva\nDuring his work at Imperva, security researcher Ron Massas uncovered a number of fascinating vulnerabilities and hacks in a number of large companies.\nIn the lecture, he will take us on a fascinating journey in the world of offensive security, and will present challenges discovered along the way and techniques that were used during the attacks.\n\nTuesday, June 4 at 18:30 at Taub 1\n\nRegister at the link\n\nEverybody is invited!
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 1
UID:eventx6a5a287eec07910601
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20240605T113000
DTEND;TZID=Asia/Jerusalem:20240605T123000
DTSTAMP;TZID=Asia/Jerusalem:20240605T113000
SUMMARY: ceClub  talk by Noga Rotman, HUJI  CE-Club: Safe Deployment of Deep Learning Solutions for Rate Control Problems  at 2024-06-05 11:30:00
DESCRIPTION:In recent years, machine learning (ML) has fueled extraordinary advances in fields such as natural language processing and computer vision. These results have encouraged researchers to utilize ML in additional application domains, such as computer and networked systems, with varying degrees of success. In this talk I will focus on the prominent challenge of rate control in computer networking, and devising a practical ML toolkit for the safe deployment of deep-learning-based solutions.Noga H. Rotman is a PhD student at the Hebrew University, advised by prof. Michael Schapira. Her research focuses on the intersection between computer networks and machine learning.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture: 94673013539 & Meyer 861 	
UID:eventx6a5a287eec08a10603
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20240605T124500
DTEND;TZID=Asia/Jerusalem:20240605T134500
DTSTAMP;TZID=Asia/Jerusalem:20240605T124500
SUMMARY: Theory Seminar  talk by Ronen Shaltiel (Haifa University)  Theory Seminar: Explicit Codes for Poly-Size Circuits and Functions that are Hard to Sample on Low Entropy Distributions  at 2024-06-05 12:45:00
DESCRIPTION:In the talk I will survey a research direction which combines coding-theory and computational complexity. This direction was introduced by Lipton, and refined by Guruswami and Smith. The goal is to construct error correcting codes which correct errors that are introduced by computationally bounded channels. This is in contrast to Hamming’s standard model which assumes a bound on the fraction of bits that a channel may flip, but allows the channel to be computationally unbounded.In recent work with Jad Silbak we give explicit constructions of codes with rate that beats the best possible rate of codes in Hamming’s standard model and matches the capacity of Shannon’s binary symmetric channels. These results are achieved against channels with small space, and channels that can be implemented by poly-size circuits (under a suitable hardness assumption).My plan is to explain the model, survey the new results, and try to show some of the components that are used in these works.Joint work with Jad Silbak.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Amado 719 
UID:eventx6a5a287eec09810604
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20240609T143000
DTEND;TZID=Asia/Jerusalem:20240609T153000
DTSTAMP;TZID=Asia/Jerusalem:20240609T143000
SUMMARY: MSC  talk by Hadas Abraham  Covering All Bases: The Next Inning in DNA Sequencing Efficiency  at 2024-06-09 14:30:00
DESCRIPTION:DNA emerges as a promising medium for the exponential growth of digital data due to its density and durability. This study extends recent research by addressing the coverage depth problem in practical scenarios, exploring optimal error-correcting code pairings with DNA storage systems to minimize coverage depth. Conducted within random access settings, the study provides theoretical analyses and experimental simulations to examine the expectation and probability distribution of samples needed for files recovery.\n
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture:99533009512 & room 601
UID:eventx6a5a287eec0a610597
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20240610T150000
DTEND;TZID=Asia/Jerusalem:20240610T160000
DTSTAMP;TZID=Asia/Jerusalem:20240610T150000
SUMMARY: pixel-club  talk by Adi Vainiger, The Andrew and Erna Viterbi Faculty of Electrical & Computer Engineering  Pixel-Club:Volumetric Signal Recovery and Calibration in Atmospheric Lidar  at 2024-06-10 15:00:00
DESCRIPTION:Atmospheric lidars are important remote sensing tools in aerosols and climate research. A pulsed time-of-flight lidar continuously samples vertical atmospheric profiles, through day and night, yielding a spatiotemporal atmospheric map. However, lidar analysis is challenged by low signal-to-noise ratios, sunlight interference, and need for frequent calibration. We advance lidar analysis. We develop a framework for simulating realistic spatiotemporal lidar data under diverse atmospheric and system conditions. This reveals limitations and flaws in standard processing pipelines. To counter that, we develop maximum likelihood estimation tailored to lidar. This significantly enhances lidar analysis, particularly during the daytime, enabling more frequent and accurate retrievals. Additionally, we develop lidar calibration based on learning, based on spatiotemporal meteorological and lidar data. This helps address low signal-to-noise ratios and yield keeps calibration more updated than the current operational approach.\n\nPh.D. Under the supervision of Prof. Yoav Schechner.\n
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Room 1061, EE Meyer Building 
UID:eventx6a5a287eec0b610607
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20240616T110000
DTEND;TZID=Asia/Jerusalem:20240616T120000
DTSTAMP;TZID=Asia/Jerusalem:20240616T110000
SUMMARY: MSC  talk by Yarin Bar  Protected Test-Time Domain Adaptation via Online Entropy Matching  at 2024-06-16 11:00:00
DESCRIPTION:In this paper, we present a novel approach for test-time domain adaptation via online self-training, consisting of two components. First, we introduce a statistical framework that detects distribution shifts in the classifier's entropy values obtained on a stream of unlabeled samples. Second, we devise an online adaptation mechanism that utilizes the evidence of distribution shifts captured by the detection tool to dynamically update the classifier's parameters. The resulting adaptation process drives the distribution of \ntest entropy values obtained from the self-trained classifier to match those of the source domain. This approach departs from the conventional self-training method, which focuses on minimizing the classifier's entropy. Our approach combines concepts in betting martingales and online learning to form a detection tool capable of quickly reacting to distribution shifts. We then reveal a tight relation between our adaptation scheme and optimal transport, which forms the basis of our novel self-supervised loss. Experimental results demonstrate that our approach improves test-time accuracy under distribution shifts while maintaining accuracy and calibration in their absence, outperforming leading entropy minimization methods across various scenarios.\n
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 9
UID:eventx6a5a287eec0c510608
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20240616T130000
DTEND;TZID=Asia/Jerusalem:20240616T140000
DTSTAMP;TZID=Asia/Jerusalem:20240616T130000
SUMMARY: MSC  talk by Akiva Michael Block  Curvature Shift in Neural Networks  at 2024-06-16 13:00:00
DESCRIPTION:Neural network optimization is a challenging task, in large part because of the nonconvexity, in general, of the loss function. Various algorithms have been proposed to solve this task, but most of them focus on the convex subspaces of the loss function to avoid the challenge of finding the optimal step in concave subspaces. In addition to avoiding the concave spaces entirely (which can bear a significant computational burden!), they also simply take the minimizer of a second-degree Taylor approximation of the loss function for their step, without addressing the prospect of a significant gap between the loss function and its second-order approximation. In our work, we tackle this problem by making use of the Lipschitz continuity of the loss function to develop the regret-optimal step in all subspaces, whether convex or concave, and discuss the validity and (sometimes) experimental success of such optimizers. We hope that our work will provide practitioners with new tools for selecting neural network training algorithms.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture: 94919856641 & Taub 9	
UID:eventx6a5a287eec0d110606
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20240616T143000
DTEND;TZID=Asia/Jerusalem:20240616T153000
DTSTAMP;TZID=Asia/Jerusalem:20240616T143000
SUMMARY: MSC  talk by Ishai Salgado  Quantitative Semantics on Jumping Finite Automata  at 2024-06-16 14:30:00
DESCRIPTION:Jumping automata are finite automata that read their input in a non-sequential manner, by allowing a reading head to “jump” between positions on the input, consuming a permutation of the input word. We argue that allowing the head to jump should incur some cost. To this end, we propose three quantitative semantics for jumping automata, whereby the jumps of the head in an accepting run define the cost of the run. The three semantics correspond to different interpretations of jumps: the absolute distance semantics counts the distance the head jumps, the reversal semantics counts the number of times the head changes direction, and the Hamming distance measures the number of letter-swaps the run makes.\nWe study these measures, with the main focus being the boundedness problem: given a jumping automaton, decide whether its (quantitative) languages are bounded by some given number k. We establish the decidability and complexity for this problem under several variants.\n
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture: 94576770275 &  Taub 301
UID:eventx6a5a287eec0dc10605
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20240619T130000
DTEND;TZID=Asia/Jerusalem:20240619T140000
DTSTAMP;TZID=Asia/Jerusalem:20240619T130000
SUMMARY: Theory Semina  talk by Omri Weinstein (Hebrew University)  Theory Seminar: Approximate Matrix Multiplication via Spherical Convolutions  at 2024-06-19 13:00:00
DESCRIPTION:We develop a new framework, Polyform, for fast approximate matrix multiplication (AMM), through sums of sparse polynomial multiplications, using (variants of) FFT. Using this framework, we obtain new data-dependent speed-accuracy tradeoffs, which often improve on the worst-case accuracy of randomized sketching algorithms. Meanwhile, Polyform can be viewed as a cheap alternative bilinear operator to matrix multiplication in Deep Neural Networks, which is our motivating application.&nbsp;Our algorithm involves unexpected connections to Additive Combinatorics and Spherical Harmonics. The former one relates the error of Polyform to the purely existential problem of finding Sumsets which minimize intersections with arithmetic progressions. In the latter, we show that optimizing the polynomial’s coefficients leads to a low-rank SDP problem generalizing Spherical Codes, and provide a novel projection-free approximation algorithm for this problem, in close to input-sparsity time ~O(nnz(A,B)). Meanwhile, our experiments demonstrate that, when using SGD to optimize the coefficients of Polyform in stat-of-art Transformer models (ViT), our algorithm provides a significant reduction in FLOPs (x2-x5) compared to vanilla MM, with little to no accuracy loss.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Amado 619
UID:eventx6a5a287eec0e710610
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20240620T080000
DTEND;TZID=Asia/Jerusalem:20240621T140000
DTSTAMP;TZID=Asia/Jerusalem:20240620T080000
SUMMARY: CSpecial Event  The Annual Faculty Hackathon is Underway! CS Hackathon - Doing Good  at 2024-06-20 08:00:00
DESCRIPTION:The annual Faculty Hackathon is underway! CS Hackathon - Doing Good\nReserve the dates: June 20-21, in Taub.\n\nThis year, in accordance with the order of the day, we will focus on developing technological solutions to increase mental resilience in cooperation with the Ministry of Defense's rehabilitation department, associations and initiatives.\nOpening of registration, development ideas, and preparation workshops - very soon.\n\nFor details and more information on the hackathon website here\n\nStay tuned!
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub Building
UID:eventx6a5a287eec0f210593
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20240620T163000
DTEND;TZID=Asia/Jerusalem:20240620T173000
DTSTAMP;TZID=Asia/Jerusalem:20240620T163000
SUMMARY: MSC  talk by Noam Koren  Multivariate Time Series Prediction  at 2024-06-20 16:30:00
DESCRIPTION:Multivariate time series forecasting is a pivotal task in several domains, including financial planning, medical diagnostics, and climate science. This paper presents the Neural Fourier Transform (NFT) algorithm, which combines multi-dimensional Fourier transforms with Temporal Convolutional Network layers to improve both the accuracy and interpretability of forecasts. The Neural Fourier Transform is empirically validated on fourteen diverse datasets, showing superior performance across multiple forecasting horizons and lookbacks, setting new benchmarks in the field. This work advances multivariate time series forecasting by providing a model that is both interpretable and highly predictive, making it a valuable tool for both practitioners and researchers. The code for this study is publicly available https://github.com/2noamk/NFT.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture: 98850618360 & Taub 601
UID:eventx6a5a287eec0fd10609
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20240625T113000
DTEND;TZID=Asia/Jerusalem:20240625T123000
DTSTAMP;TZID=Asia/Jerusalem:20240625T113000
SUMMARY: pixel-club  talk by Roi Ronen  Pixel Club: Robust and Fast Volumetric Tomography of Natural Objects for Climate Studies  at 2024-06-25 11:30:00
DESCRIPTION:Computed tomography (CT) aims to recover the volumetric three dimensional (3D) structure of heterogeneous objects. Traditionally, CT refers to a medical imaging modality. There the object, radiation source and detector array are fully controlled. This results in a linear image formation model. In contrast, we seek CT of natural objects, acquired outdoors, in an uncontrolled environment. We focus on underwater plankton and cloud droplets, which strongly affect climate. Cloud imaging is a highly complex and recursive model, involving solar radiation multiple scattering which makes CT non-linear in the unknowns. Moreover, cloud CT requires multi-view images acquired by satellites, but clouds evolve while the satellites overfly. Hence, we derive spatiotemporal CT of time-varying clouds. Moreover, we accelerate scattering-CT analysis by several orders of magnitude: We develop a deep neural network for variable imaging projection cloud tomography (VIP-CT). VIP-CT is agnostic by construction to the cameras’ positions, being flexible in imaging geometry. Then, we extend VIP-CT to estimate per 3D cloud location a function:. Consequently, 3D recovery includes various statistics, such as the most probable result and uncertainty. Also, we show the recovery uncertainty effect on precipitation and renewable energy forecasts. To improve out-of-distribution inference, we incorporate a novel self-supervised learning through differential rendering. The ability to do CT in variable geometries is further generalized for underwater plankton. There, each specimen imaged only once, at random unknown pose and scale. Using this image ensemble, we achieve plankton 3D tomography and estimate population statistics. To counter errors due to plankton non-rigid deformations, we weigh the data by an advanced statistical model, developed for Cryo-EM.Roi Ronen is a Ph.D. student under the supervision of Prof. Yoav Schechner and a researcher at Amazon Web Service (AWS) – AI labs. His research explores the interface between machine learning, physics-based rendering, 3D tomography and computational photography. His paper won the JOSA A Emerging Researcher Best Paper Prize for 2021.Roi also received the Jewish National Fund Climate scholarship.\nPersonal web page\n
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom lecture &amp; 1061, Meyer Building
UID:eventx6a5a287eec10810613
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20240626T113000
DTEND;TZID=Asia/Jerusalem:20240626T123000
DTSTAMP;TZID=Asia/Jerusalem:20240626T113000
SUMMARY: ceClub  talk by Gal Assa  CE Club: Persistent Memory Programming without Persistent Memory  at 2024-06-26 11:30:00
DESCRIPTION:The arrival of persistent memory devices to consumer market has revived the interest in transactional durable algorithms. Persistent memory is touted as having two attributes that distinguish it from other storage technologies: fine access granularity and fast persistece.In the first part of the talk we investigate how these attributes differentiate persistent memory from block storage in the context of buffered durability – a relaxed approach that allows some progress loss upon system crash. We present TL4x, a novel persistent transactional memory framework capable of providing buffered durable linearizable transactions with high scalability for disjoint writes and efficient persistence on either persistent memory or block storage devices. TL4x maintains a volatile snapshot which is used for both persistence and irrevocable read-only transactions, which can facilitate long range-query operations.The second part of the talk will be dedicated to CXL0, the first programming model for disaggregated memory over the newly emerging CXL standard. It addresses memory sharing between processors with different architectures in a cache coherent manner. We will discuss the need for a new programming model and present a transformation that provides linearizable algorithms with durable linearizability under a partial-system crash failure.Gal is a PhD student supervised by Prof. Idit Keidar.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture: 94673013539 & Zisapel 506 	
UID:eventx6a5a287eec11610615
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20240626T130000
DTEND;TZID=Asia/Jerusalem:20240626T140000
DTSTAMP;TZID=Asia/Jerusalem:20240626T130000
SUMMARY: Theory Seminar  talk by Tomer Even (Technion)  Theory Seminar: Fast Approximate Counting of Cycles  at 2024-06-26 13:00:00
DESCRIPTION:We consider the problem of approximate counting of triangles and longer fixed length cycles in undirected and directed graphs. We provide an algorithm which is faster as t, the number of copies of the searched subgraph, increases. Our running time, which significantly improves upon the state of the art (Tětek [ICALP’22]), is the same as that of multiplying an n x (n/t) matrix by an (n/t) x n matrix, up to  polylogarithmic factors. Finally, we show that under popular fine-grained hypotheses, this running time is optimal.\n\nThe talk is based on a joint work with Keren Censor-Hillel and Virginia Vassilevska Williams. To appear in ICALP 2024.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Amado 719 
UID:eventx6a5a287eec12110618
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20240627T100000
DTEND;TZID=Asia/Jerusalem:20240627T110000
DTSTAMP;TZID=Asia/Jerusalem:20240627T100000
SUMMARY: MSC  talk by Sally Turutov  Advancing Molecular Frontiers: Redefining Optimization Goals in Drug Design  at 2024-06-27 10:00:00
DESCRIPTION:Deep learning has revolutionized drug development by enhancing various stages of the process, yet traditional optimization goals often lack the sophistication required for real-world applications. This work tackles more complex and nuanced problems beyond conventional property enhancement, focusing on optimizing under patentability constraints and translating preclinical success in animals to human clinical trials. In addressing patentability, we introduce a patent loss mechanism and the Molecular Optimization Model with Patentability Constraint (MOMP) to ensure generated molecules are novel and sufficiently different from existing patents, thereby enhancing their potential for successful patenting. Our approach navigates the intricate landscape of patentability, diverging from typical property optimization methods that rely on straightforward classifier definitions.\nAdditionally, we address the challenge of translating drug efficacy from animals to humans, introducing the Biological Complexity Curriculum Learning for Molecular Activity Prediction in Humans (BioMolX). Utilizing state-of-the-art Graph Neural Networks and curriculum learning, BioMolX progressively trains on animal data of increasing physiological complexity to enhance predictive accuracy for human tissues. Building on this, our Rat2Human model optimizes molecules effective in rats to ensure similar efficacy in humans, employing a transformer architecture to learn transformations at both atomic and molecular levels across multiple tissues. These innovative methodologies demonstrate a holistic approach to drug development, tackling complex challenges to generate novel compounds and accelerate the delivery of effective treatments to patients.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture
UID:eventx6a5a287eec12c10614
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20240630T103000
DTEND;TZID=Asia/Jerusalem:20240630T113000
DTSTAMP;TZID=Asia/Jerusalem:20240630T103000
SUMMARY: ceClub  talk by Rotem Ben Hur  ceClub: Enhancing Computer Performance with memristive Memory Processing Units: From General-Purpose Automation to DNA Sequencing Acceleration  at 2024-06-30 10:30:00
DESCRIPTION:Computer systems that facilitate tight integration of data storage and processing can eliminate the "memory wall" and "power wall" bottlenecks. The memristive Memory Processing Unit (mMPU) architecture contains memory cells that can also execute logical functions, such as Memristor-Aided Logic (MAGIC) NOR gates. The mMPU supports both general-purpose computing and application-specific acceleration. This research contributes to both these aspects of the mMPU.First, we introduce frameworks that automate the mapping of any desired logical function into the mMPU. These frameworks, called SIMPLE and SIMPLER, employ algorithms that optimize any arbitrary in-memory MAGIC operation sequences. Various performance criteria can be configured, including latency (SIMPLE), throughput (SIMPLER), chip area, and energy consumption. Both frameworks provide significant improvements over previous solutions.Focusing on the specific application of DNA sequencing, we present an mMPU-based accelerator called DART-PIM. Unlike prior studies which suffer from data-transfer bottlenecks, DART-PIM integrates all stages on a single chip for maximal end-to-end performance. This is achieved through a unique data organization technique and a novel algorithmic flow tailored for in-memory implementation. DART-PIM achieves two orders of magnitude improvement in throughput and energy efficiency, compared to state-of-the-art existing architectures.Rotem is a PhD student supervised by Prof. Shahar Kvatinsky.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture: 92829605892 & Zisapel 608  	
UID:eventx6a5a287eec13810622
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20240630T143000
DTEND;TZID=Asia/Jerusalem:20240630T153000
DTSTAMP;TZID=Asia/Jerusalem:20240630T143000
SUMMARY: MSC  talk by Dvir ben Shabat  GradHC: Highly reliable Clustering Algorithm for DNA Storage Systems  at 2024-06-30 14:30:00
DESCRIPTION:As data storage challenges grow and existing technologies approach their limits, synthetic DNA emerges as a promising storage solution due to its remarkable density and durability advantages. While cost remains a concern, emerging sequencing and synthetic technologies aim to mitigate it, yet introduce challenges such as errors in the storage and retrieval process. One crucial task in a DNA storage system is clustering numerous DNA reads into groups that represent the original input strands. We will review different methods for evaluating clustering algorithms and introduce a novel clustering algorithm for DNA storage systems, named Gradual Hash-based clustering (GradHC). The primary strength of GradHC lies in its capability to cluster with excellent accuracy various types of designs, including varying strand lengths, cluster sizes (including extremely small clusters), and different error ranges. Benchmark analysis demonstrates that GradHC is significantly more stable and robust than other clustering algorithms previously proposed for DNA storage, while also producing highly reliable clustering results.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture: 5960823195
UID:eventx6a5a287eec15a10611
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20240630T143000
DTEND;TZID=Asia/Jerusalem:20240630T153000
DTSTAMP;TZID=Asia/Jerusalem:20240630T143000
SUMMARY: ceClub  talk by Orian Leitersdorf  ceClub: Advancing Computer Science with Digital Processing in Memory  at 2024-06-30 14:30:00
DESCRIPTION:The memory wall bottleneck is throttling the performance of data-intensive applications as the data transfer between the processing units (e.g., CPU, GPU cores) and the memory is significantly slower than the compute itself. Therefore, emerging digital processing-in-memory systems overcome this bottleneck by performing parallel bitwise logic within the memory arrays themselves. This enables a drastic reduction in data transfer for vectored operations since the same instruction may be repeated across the entire vector dimension, thereby leading to the effectiveness of PIM for applications that require high-throughput vectored arithmetic.In this seminar, we explore the extension of these high-throughput bitwise operations to reliable large-scale applications. We begin by constructing high-throughput arithmetic for both fixed-point and floating-point numbers based on the supported parallel bitwise operations, incorporating a variety of techniques to convert the control flow of floating-point operations into data-flow. We continue by demonstrating the effectiveness of these parallel vectored arithmetic operations for the acceleration of large scale applications such as matrix multiplication and Fast Fourier Transform (FFT), and propose an end-to-end programming framework that automatically converts high-level python instructions to underlying bitwise operations. Lastly, we explore algorithmic techniques that aim to improve the reliability of the memory and logic, thereby compensating for soft errors.Orian is a PhD student supervised by Prof. Shahar Kvatinsky.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture: 93857585105 & Zisapel 608  	
UID:eventx6a5a287eec16c10623
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20240630T183000
DTEND;TZID=Asia/Jerusalem:20240630T203000
DTSTAMP;TZID=Asia/Jerusalem:20240630T183000
SUMMARY: CSpecial Event  Come be Part of the Faculty's CTF Group - Meeting on June 30th  at 2024-06-30 18:30:00
DESCRIPTION:Come be part of the Faculty's Capture The Flag - CTF group!!\n\nThe faculty CTF group, Technipwn, holds bi-weekly meetings where we solve challenges, train for competitions and have fun\nThe meetings are suitable for both beginners and experienced participants, and are held on Sundays at Taub 9.\n\nOrganized and managed by students, and intended for all students (not only from computer science!), with and without CTF experience. The sessions include practical experience in solving challenges.\n\nThe next meeting will be held on Sunday, June 30, where we will participate as a group at UIUCTF, everyone is invited!\n\nFor questions and answers: https://technionctf.com/faq\n\nOur WhatsApp group: https://chat.whatsapp.com/BC5nhbQhlhv4NoOBVStjET\n\nwaiting for you!!
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 9
UID:eventx6a5a287eec17810621
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20240701T140000
DTEND;TZID=Asia/Jerusalem:20240701T150000
DTSTAMP;TZID=Asia/Jerusalem:20240701T140000
SUMMARY: PHD  talk by Mohammad Agbarya  What Can be Learned from Mixing Memory Pages of Different Sizes  at 2024-07-01 14:00:00
DESCRIPTION:Cycle-accurate simulations, frequently used by computer architects, incur substantial overheads. To mitigate this, recent virtual memory studies have adopted a lighter-weight methodology that leverages partial simulations of only the memory subsystem. This approach feeds simulation outputs into a mathematical linear model to predict execution runtimes. While this methodology accelerates the simulation process, its accuracy has traditionally been assumed rather than rigorously validated.In this study, we challenge this assumption with the development of Mosalloc, the Mosaic Memory Allocator. Mosalloc supports the virtual memory of applications with a variety of page sizes, specifically 4KiB, 2MiB, and 1GiB, creating diverse \\\"mosaic\\\" memory layouts. Contrary to previous approaches that utilized a singular page size to generate merely two execution samples for model development, Mosalloc is capable of producing a broad spectrum of samples, thereby allowing empirical validation of model accuracy. Our evaluations reveal that existing models exhibit prediction errors ranging from 25% to 192%. In response, we propose Mosmodel, a new runtime model that confines the maximal cross-validation error to below 6%, significantly enhancing reliability for experimental exploration.The efficacy of these models hinges on the availability of diverse memory layouts. As the number of feasible layouts escalates exponentially with address space size, the selection process becomes increasingly complex. To address this, we introduce Moselect, an algorithm that autonomously identifies optimal memory layouts that ensure up to 4% gaps between consecutive data points, averaging 46 layouts. Moselect not only augments model accuracy with its diverse samples but also simplifies the modeling and simulation processes. This simplification trades a marginal increase in maximal error of up to 2.5% for significant reductions in budget and execution time. Moreover, it facilitates the use of a TLB-only simulator, eliminating the need for comprehensive cache hierarchy simulations and thus minimizing development efforts.By amalgamating Mosalloc, Moselect, and Mosmodel, we present a comprehensive framework that automates memory layout selection, benchmark execution, and accurate runtime model construction. This framework also introduces techniques to streamline operations without materially compromising accuracy. Collectively, these innovations significantly advance the state-of-the-art in virtual memory research, providing a robust platform for future exploration and development.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture: 91705525071 & Taub 601
UID:eventx6a5a287eec18410625
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20240702T143000
DTEND;TZID=Asia/Jerusalem:20240702T153000
DTSTAMP;TZID=Asia/Jerusalem:20240702T143000
SUMMARY: colloq  talk by Prof. Oded Margalit (Ben Gurion University)   Wrong Mathematical Proofs for Possibly Correct Claims: Implications and Applications in Cyber Security and Chip Design  at 2024-07-02 14:30:00
DESCRIPTION:In this talk, we'll explore intriguing parallels between incorrect mathematical proofs and critical failures in technology, demonstrating their impact and lessons learned, like:\n1. Mathematical Proofs and Social Engineering: How hand-waving a mathematical proof is similar to social engineering, and why larger hardware implementations might surprisingly be more efficient.\n2. Cybersecurity and Mathematical Proofs: The surprising connection between a cyber attack that nearly caused an airplane crash and the proof that any finite set of horses has the same color, illustrating the importance of thorough proof validation in ensuring safety.\n3. Hacking games and testing: Drawing a parallel between winning Super Mario in less than two minutes and the four-color theorem, highlighting similar challenges that cost Intel $475M.\n\nBio: Oded Margalit is the Head Scientist at NextSilicon and an Adjunct Professor in the Computer Science department at Ben-Gurion University of the Negev. Previously, he was a researcher at the IBM Haifa Research Lab (HRL) and the CTO of Citi's Cyber Security Innovation Center (CSIC). Oded has an extensive background in machine learning, optimization, formal verification, and cybersecurity.\nHe is also dedicated to educational outreach, having created mathematical and programming challenges for a wide range of learners, including elementary school students (CodeGuru), high school students (CodeGuru Xtreme), university students (IEEEXtreme), and other audiences (PonderThis and more).\n\nTechnion Host: Erez Petrank 
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub Auditorium 012
UID:eventx6a5a287eec19210617
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20240702T170000
DTEND;TZID=Asia/Jerusalem:20240702T200000
DTSTAMP;TZID=Asia/Jerusalem:20240702T170000
SUMMARY: CSpecial Event  Technipwn Group flies you to Vegas!  at 2024-07-02 17:00:00
DESCRIPTION:Technipwn Group flies you to Vegas!This August, the faculty CTF group is participating in the DEFCON conference in Las Vegas and invites you to join!This coming tuesday we will hold a competition aimed at choosing the team that will fly to represent us at the conference.The topics of the competition are the usual topics of the ctf competitions such as: web, reverse, crypto, pwn and more!You are invited to prepare and come to compete on Tuesday 7/2 at 17:00-20:00Register at the link
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 9
UID:eventx6a5a287eec19e10627
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20240703T110000
DTEND;TZID=Asia/Jerusalem:20240703T120000
DTSTAMP;TZID=Asia/Jerusalem:20240703T110000
SUMMARY: MSC  talk by Almog David  Sequential Signal Mixing Aggregation For Message Passing Graph Neural Networks  at 2024-07-03 11:00:00
DESCRIPTION:Message Passing Graph Neural Networks (MPGNNs) have emerged as the standard method for modeling complex interactions across diverse graph entities. While the theory of such models is widely investigated, their aggregation module has not received sufficient attention. Sum-based aggregators have solid theoretical foundations regarding their separation capabilities. However, practitioners often prefer using more complex aggregations and mixtures of diverse aggregations.\nIn this research, we unveil a possible explanation for this gap. We claim that sum-based aggregators fail to "mix" features belonging to distinct neighbors, preventing them from succeeding at downstream tasks. To this end, we introduce Sequential Signal Mixing Aggregation (SSMA), a novel plug-and-play aggregation for MPGNNs. SSMA treats the neighbor features as 2D discrete signals and sequentially convolves them, inherently enhancing the ability to mix features attributed to distinct neighbors. By performing extensive experiments, we show that combining SSMA with well-established MPGNN architectures achieves substantial performance gains across various benchmarks, achieving new state-of-the-art results in many settings.\n
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Room 014
UID:eventx6a5a287eec1a910616
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20240709T113000
DTEND;TZID=Asia/Jerusalem:20240709T123000
DTSTAMP;TZID=Asia/Jerusalem:20240709T113000
SUMMARY: pixel-club  talk by Roy Ganz - The Andrew and Erna Viterbi Faculty of Electrical & Computer Engineering  Pixel Club - On The Benefits Of Models With Perceptually Aligned Gradients  at 2024-07-09 11:30:00
DESCRIPTION:Deep learning has revolutionized computer vision, achieving unprecedented performanc in tasks like classification and detection. However, these models are highly susceptible to adversarial attacks, prompting the development of robust training methods. A notable outcome of such training is the phenomenon of Perceptually Aligned Gradients (PAG), where input gradients align semantically with human perception. Our research explores both the practical and theoretical implications of PAG. We introduce BIGRoC, a modelagnostic image refinement method that leverages PAG to enhance generated images from any source. Additionally, we study the connection between PAG and adversarial robustness, demonstrating that the connection is bidirectional. Finally, we extend PAG research to the multimodal vision-language domain, unveiling CLIPAG, a CLIP-based model that improves generative tasks and achieves text-to-image generation without a traditional generator.\nPh.D. Under the supervision of Prof. Michael Elad.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture: 94860001040
UID:eventx6a5a287eec1b410632
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20240710T123000
DTEND;TZID=Asia/Jerusalem:20240710T133000
DTSTAMP;TZID=Asia/Jerusalem:20240710T123000
SUMMARY: PHD  talk by Ido Galil  Large Scale Studies Of Deep Neural Networks For Uncertainty Estimation And Class-out-of-Distribution Detection  at 2024-07-10 12:30:00
DESCRIPTION:In this seminar, I will discuss two of my papers (published in ICLR 2023) on uncertainty estimation and class-out-of-distribution detection. We present a novel framework to benchmark the ability of image classifiers to detect class-out-of-distribution (C-OOD) instances (i.e., instances whose true labels do not appear in the training distribution) at various levels of detection difficulty. We apply this technique to ImageNet, and benchmark 500+ pretrained, publicly available, ImageNet-1k classifiers. We then evaluate these classifiers both for their C-OOD detection performance (second paper) and for their uncertainty estimation performance (first paper), i.e., their ranking, calibration, and selective classification performance. This results in a large-scale study that reveals many factors (such as architecture types and training regimes) previously unknown to contribute to performance.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture: 92810808120 & Taub 9
UID:eventx6a5a287eec1bf10619
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20240710T123000
DTEND;TZID=Asia/Jerusalem:20240710T143000
DTSTAMP;TZID=Asia/Jerusalem:20240710T123000
SUMMARY: CSpecial Event  Spotlight Day - Nvidia  at 2024-07-10 12:30:00
DESCRIPTION:First spotlight day of the semester!!\n\nNvidia is coming to meet you in Taub\n\nWednesday, July 10, between 12:30-2:30 PM in the Taub lobby\n\nIn the program: a meeting with the recruitment team, the engineers\n\nwaiting for you!!
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:CS Taub Lobby
UID:eventx6a5a287eec1ca10624
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20240710T123000
DTEND;TZID=Asia/Jerusalem:20240710T140000
DTSTAMP;TZID=Asia/Jerusalem:20240710T123000
SUMMARY: colloq  talk by Chen Shani (Stanford)  Designing Language Models to Think Like Humans  at 2024-07-10 12:30:00
DESCRIPTION:While language models (LMs) show impressive text manipulation capabilities, they also lack commonsense and reasoning abilities and are known to be brittle. In this talk, I will suggest a different LMs design paradigm, inspired by how humans understand it. I will present two papers, both shedding light on human-inspired NLP architectures aimed at delving deeper into the meaning beyond words. \n \nThe first paper [1] accounts for the lack of commonsense and reasoning abilities by proposing a paradigm shift in language understanding, drawing inspiration from embodied cognitive linguistics (ECL). In this position paper we propose a new architecture that treats language as inherently executable, grounded in embodied interaction, and driven by metaphoric reasoning. \n \nThe second paper [2] shows that LMs are brittle and far from human performance in their concept-understanding and abstraction capabilities. We argue this is due to their token-based objectives, and implement a concept-aware post-processing manipulation, showing it matches human intuition better. We then pave the way for more concept-aware training paradigms. \n \n \n[1] Language (Re)modelling: Towards Embodied Language Understanding\nRonen Tamari, Chen Shani, Tom Hope, Miriam R L Petruck, Omri Abend, and Dafna Shahaf. 2020. \nIn Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics (ACL), pages 6268–6281, Online. Association for Computational Linguistics.\n \n[2] Towards Concept-Aware Large Language Models\nShani, Chen, Jilles Vreeken, and Dafna Shahaf. \nIn Findings of the Association for Computational Linguistics: EMNLP 2023, pp. 13158-13170. 2023.\n \n \nBio: Chen Shani is a post-doctoral researcher at Stanford's NLP group, collaborating with Prof. Dan Jurafsky. Previously, she pursued her Ph.D. at the Hebrew University under the guidance of Prof. Dafna Shahaf and worked at Amazon Research. Her focus lies at the intersection of humans and NLP, where she implements insights from human cognition to improve NLP systems.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 337
UID:eventx6a5a287eec1d410628
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20240710T130000
DTEND;TZID=Asia/Jerusalem:20240710T140000
DTSTAMP;TZID=Asia/Jerusalem:20240710T130000
SUMMARY: Theory Semina  talk by Shai Ben-David (University of Waterloo)   Theory Seminar: Learning Probability Distributions; What Can, What Can't Be Done  at 2024-07-10 13:00:00
DESCRIPTION:Characterizing learnability by a combinatorial dimension is a hallmark of machine learning theory. Starting from the fundamental characterization of binary classification PAC by the VC-dimension, through the characterization of online mistake bound learnability by the Littlestone dimension, all the way to recent papers characterizing multi-class learning and differential privacy.\nHowever, for some basic learning setup no such characterizations have been provided.\nI will focus on one of these setups &mdash; unsupervised learning probability distributions.\nI will start by showing that with no prior assumptions density estimation cannot have success guarantees.\nNext, I will describe a positive learnability result &mdash; settling the question of the sample complexity of learning mixtures of Gaussians.\nFinally, I will turn to the question of characterizing learnable classes of distributions.\nI will show that there can be no combinatorial characterization of the family of learnable classes of distributions, as well as similar impossibility results for other learnability setups (some of these going back to work of Giora Benedek and Alon Itai). This settles open problems that have been open for more than 30 years.\nThe first part is based on joint work with Hasan Ashiani, Nick Harvey, Chris Law, Abas Merhabian and Yaniv Plan and the second part on work with my student Tosca Lechner.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Amado 814
UID:eventx6a5a287eec1e110633
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20240711T163000
DTEND;TZID=Asia/Jerusalem:20240711T173000
DTSTAMP;TZID=Asia/Jerusalem:20240711T163000
SUMMARY: MSC  talk by Ori Mazor  Interactions between Blockchain Networks Based on Market Analysis  at 2024-07-11 16:30:00
DESCRIPTION:Cryptocurrencies have gained popularity in recent years. They are often implemented on one or more of the over 1000 existing blockchain networks, including the popular Bitcoin and Ethereum networks. Recently, various technologies known as blockchain interoperability have been developed to connect these different blockchains, creating an interconnected blockchain ecosystem. Interoperability refers to the ability of blockchains to share information with each other. Decentralized Exchanges (DEXs), which are peer-to-peer marketplaces where traders can exchange cryptocurrencies, play a significant role in this interconnected ecosystem. We aim to understand the blockchain ecosystem and the connections between blockchains, we view that as the blockchain interoperability graph. Our approach is based on analyzing the correlation between cryptocurrency prices implemented over different blockchains. We examine over 4800 cryptocurrencies implemented on 76 blockchains based on their daily prices. This experimental study has potential implications for decentralized finance (DeFi), including portfolio investment strategies and risk management. Additionally, we present a framework to study cross-chain arbitrage between DEXs. We demonstrate the framework by analyzing two popular DEXs: QuickSwap which is implemented on the Polygon blockchain network and PancakeSwap that is implemented on the BNB Chain blockchain network. We study the potential revenue from conducting cross-chain arbitrage and lay the basis for understanding its potential impact on the future of blockchain technology. Overall, the analysis provides insight into the evolving blockchain ecosystem, highlighting opportunities for innovation and financial strategies in the decentralized market.\n
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture: 5701477766 	
UID:eventx6a5a287eec1ed10626
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20240716T113000
DTEND;TZID=Asia/Jerusalem:20240716T123000
DTSTAMP;TZID=Asia/Jerusalem:20240716T113000
SUMMARY: pixel-club  talk by Niv Cohen  Discovering and Erasing Unsafe Concepts  at 2024-07-16 11:30:00
DESCRIPTION:\n\n\n\nThe rapid growth of generative models allows an ever-increasing variety of capabilities. Yet, these models may also produce undesired content such as unsafe images, private information, or copyrighted material.\nIn this talk, I will discuss practical methods to prevent undesired generation and the evaluation of such methods. First, I will show how the challenge of avoiding undesired generations manifested itself in a simple Capture-the-Flag LLM setting, where even our top defense strategy was breached. Next, I will demonstrate a similar vulnerability in state-of-the-art concept erasure methods for Text-to-Image models. Finally, I will describe the notion of &lsquo;Unconditional Concept Erasure&rsquo; aiming to mitigate these issues. I will show that Task Vectors can achieve Unconditional Concept Erasure, and discuss the opportunities and limitations of applying Task Vectors in practice.\n\n\n\n\nNiv is a postdoctoral researcher at New York University hosted by Prof. Chinmay Hegde. He received a BSc. in mathematics with physics in the Technion Excellence Program. He received a Ph.D. in computer science from the Hebrew University of Jerusalem, advised by Prof. Yedid Hoshen. Niv was awarded the Israeli data science scholarship for outstanding postdoctoral fellows (VATAT). He is interested in model personalization, anomaly detection, and AI Safety for Language and Vision and Language models.\n&nbsp;\n\n\n\n
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Room 1061, Meyer Building
UID:eventx6a5a287eec1fa10635
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20240716T130000
DTEND;TZID=Asia/Jerusalem:20240716T140000
DTSTAMP;TZID=Asia/Jerusalem:20240716T130000
SUMMARY: MSC  talk by Dani Kogan  Hardware Sub-circuit Recognition via Subgraph Localization  at 2024-07-16 13:00:00
DESCRIPTION:Hardware Reverse Engineering (HRE) involves gate-level Netlist extraction and specification discovery, wherein graph-based methods play a crucial role in identifying sub-circuits. We propose a novel approach for Subcircuit Recognition, essential for specification discovery in HRE, leveraging existing graph similarity kernels. Focusing on the reduced problem we formulate as Subgraph Localization, we delineate two key components: graph similarity metric and graph adjacency mask inference, crucial for accurately locating subgraphs amidst complex circuit representations. Due to the dearth of domain examples, traditional supervised learning methods prove inadequate, prompting the formulation of a self-supervised approach tailored to the unique challenges of the circuit identification problem. Our contributions encompass the exploration of a novel Netlist graph representation, the formulation of an end-to-end learnable subgraph localization scheme, and the development of a comprehensive framework for evaluating graph similarity methods in this context. Moreover, we present techniques for overcoming challenges such as lack of labels and matching non-isomorphic subgraphs.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 9, https://technion.zoom.us/j/92890801845
UID:eventx6a5a287eec20910634
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20240717T093000
DTEND;TZID=Asia/Jerusalem:20240717T103000
DTSTAMP;TZID=Asia/Jerusalem:20240717T093000
SUMMARY: MSC  talk by Eitan Bloch  On the Algorithmic Foundations of Task Assistance Planning  at 2024-07-17 09:30:00
DESCRIPTION:In this work we introduce the problem of task assistance planning where we are given two robots Rtask and Rassist. The first robot, Rtask, is in charge of performing a given task by executing a precomputed path. The second robot, Rassist, is in charge of assisting the task performed by Rtask using on-board sensors. The ability of Rassist to provide assistance to Rtask depends on the locations of both robots. Since Rtask is moving along its path, Rassist may also need to move to provide as much assistance as possible. The problem we study is how to compute a path for Rassist so as to maximize the portion of Rtask&rsquo;s path for which assistance is provided. We limit the problem to the setting where Rassist moves on a roadmap which is a graph embedded in its configuration space and show that this problem is NP-hard. Fortunately, we show that when Rassist moves on a given path, and all we have to do is compute the times at which Rassist should move from one configuration to the following one, we can solve the problem optimally in polynomial time. Together with carefully crafted upper bounds, this polynomial-time algorithm is integrated into a Branch and Bound-based algorithm that can compute optimal solutions to the problem outperforming baselines by several orders of magnitude. We demonstrate our work empirically in simulated scenarios containing both planar manipulators and UR robots as well as in the lab on real robots.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture: 98269146871 &amp; Computational Robotics Lab (CRL), 1st floor, Taub building
UID:eventx6a5a287eec21410620
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20240717T130000
DTEND;TZID=Asia/Jerusalem:20240717T160000
DTSTAMP;TZID=Asia/Jerusalem:20240717T130000
SUMMARY: CSpecial Event  Google are calling to all Code Wizards  at 2024-07-17 13:00:00
DESCRIPTION:Calling all Code Wizards!\nEver wondered what it's like to be a software engineer at the heart of Google's innovation? Well, grab your coding wands and get ready to find out!Get a behind-the-scenes look on how Google connects 2 billion daily users through the OneGoogle team. Come meet the engineers behind unified identity across Google products to discover how we tackle unprecedented scale and shape the future of cross-product integration, ensuring smooth user experiences on a global level.\nBut that's not all! You'll also get:\nThe inside scoop on their personal journeys at Google\nThe chance to ask them questions - bring your A-game!&nbsp;\nInterview workshop with tips and tricks for landing that coveted Google internship or full-time role &ndash; consider this your cheat code! If you're a Computer Science / Electrical Engineering student with a passion for code and a thirst for innovation, this is your golden ticket. Don't miss out.\nWhen: Wednesday, July 17th 2024 at 13:00-14:30 Where: The Faculty of Computer Science, Technion, Taub 2 How: Space is limited so please register through the form here! &nbsp;\nHope to see you there,&nbsp;\nGoogle Student Talent Outreach Team
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 2
UID:eventx6a5a287eec22010642
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20240717T130000
DTEND;TZID=Asia/Jerusalem:20240717T140000
DTSTAMP;TZID=Asia/Jerusalem:20240717T130000
SUMMARY: Theory Semina  talk by Ariel Kulik   Improved Approximations for Vector Bin Packing via Iterative Randomized Rounding  at 2024-07-17 13:00:00
DESCRIPTION:The talk will focus on the d-dimensional vector bin packing problem. The input for the problem is a set of items, each associated with a d-dimensional weight vector. The objective is to partition the items into a minimal number of bins, such that the total weight of items in a bin is at most one in every dimension. The problem is a basic generalization of the classic Bin Packing problem and has various applications, such as virtual machines assignment in cloud environments.\nIn the talk, we will show how iterative randomized rounding, a simple and intuitive algorithmic technique, can be used to attain improved approximation for the d-dimensional vector bin packing problem and other related problems. The technique works in iterations, where each iteration solves a configuration LP relaxation for the residual instance (from previous iterations) and samples a small number of configurations based on the solution for the configuration LP. The main challenge in the analysis is to show that the value of the configuration LP decreases over time.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Amado 814
UID:eventx6a5a287eec23310640
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20240722T153000
DTEND;TZID=Asia/Jerusalem:20240722T170000
DTSTAMP;TZID=Asia/Jerusalem:20240722T153000
SUMMARY: colloq  talk by Noga Alon (Princeton and Tel Aviv University)  Distance problems for typical norms  at 2024-07-22 15:30:00
DESCRIPTION:"Noga Alon is being awarded the 2024 Wolf Prize for his profound impact on Discrete Mathematics and related areas."\nI will describe a recent joint work with Matia Bucic and Lisa Sauermann about the investigation of extremal problems in discrete geometry for typical norms.\nThe results include surprisingly tight solutions of the unit and distinct distances problems for typical norms, as well as a determination of the chromatic number of the unit distance graph for a typical planar norm.\nThis settles, in a strong form, questions and conjectures of Matousek, of Brass, of Chilakamarri and Robertson and of Brass, Moser and Pach.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Amado 232
UID:eventx6a5a287eec23f10647
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20240722T190000
DTEND;TZID=Asia/Jerusalem:20240722T220000
DTSTAMP;TZID=Asia/Jerusalem:20240722T190000
SUMMARY: CSpecial Event  "Research on the bar" night  at 2024-07-22 19:00:00
DESCRIPTION:Please register here:&nbsp;\nhttps://docs.google.com/forms/d/e/1FAIpQLSe8pYE1eVqgTK1xTf-4xyCEca4-bx12RuXtkXMcC5-4-ue_sQ/viewform&nbsp;
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub Terrace
UID:eventx6a5a287eec24910643
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20240724T110000
DTEND;TZID=Asia/Jerusalem:20240724T120000
DTSTAMP;TZID=Asia/Jerusalem:20240724T110000
SUMMARY: MSC  talk by Yanay Soker  Prediction of Model Editing Success  at 2024-07-24 11:00:00
DESCRIPTION:In LLMs, the ability to update factual knowledge of the model is an important and attractive ability, due to the need to deal with everchanging nature of information in the world.&nbsp;However, predicting whether an edit applied to a LLM will be successful or not is difficult. In this work, we suggest two metrics that can predict the editing success:&nbsp;(1) where the knowledge is stored in the parameters as reflected by the logit-lens technique;&nbsp;(2) the probability that the model assigns to the correct output.&nbsp;We find a correlation between the location of the knowledge and the optimal layer for editing, as well as between the output probability and the success. Moreover, we found a differential relation between the output probability and each component of the success.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 8, https://technion.zoom.us/j/3010716581?omn=97024786755
UID:eventx6a5a287eec25310650
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20240724T124500
DTEND;TZID=Asia/Jerusalem:20240724T134500
DTSTAMP;TZID=Asia/Jerusalem:20240724T124500
SUMMARY: Theory Semina  talk by Arnold Filtser (Bar-Ilan university)  On Sparse Partitions  at 2024-07-24 12:45:00
DESCRIPTION:A partition mathcal{P} of a metric space (X,d_X) is (sigma,tau,Delta)-sparse if each cluster has a diameter at most Delta, and every ball of radius Delta/sigma intersects at most tau clusters. In this talk, we will explore the construction and different applications of sparse partitions in their various forms over the years. As time allows, we will discuss applications to: Universal TSP, Steiner point removal, universal Steiner tree, and facility location.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Amado 814
UID:eventx6a5a287eec25e10651
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20240724T143000
DTEND;TZID=Asia/Jerusalem:20240724T153000
DTSTAMP;TZID=Asia/Jerusalem:20240724T143000
SUMMARY: MSC  talk by Tomer Cohen  The capacity and covering depth of composite DNA  at 2024-07-24 14:30:00
DESCRIPTION:This paper studies two problems that are motivated by the novel recent approach of composite DNA that takes advantage of the DNA synthesis property which generates a huge number of copies for every synthesized strand. Under this paradigm, every composite symbols does not store a single nucleotide but a mixture of the four DNA nucleotides. In the first problem, our goal is study how to carefully choose a fixed number of mixtures of the DNA nucleotides such that the decoding probability by the maximum likelihood decoder is maximized. The second problem studies the expected number of strand reads in order to decode a composite strand or a group of composite strands.&nbsp;
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Ullmann 706 https://technion.zoom.us/j/91895081808
UID:eventx6a5a287eec26810649
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20240728T183000
DTEND;TZID=Asia/Jerusalem:20240728T203000
DTSTAMP;TZID=Asia/Jerusalem:20240728T183000
SUMMARY: CSpecial Event  The CTF Technipwn group invites you to a special guest lecture from Intel  at 2024-07-28 18:30:00
DESCRIPTION:The CTF Technipwn group invites you to a special guest lecture:RingHopper - the SMM weaknesses we found in billions of devices - a security researcher from Intel An attacker who manages to penetrate the SMM mode can bypass almost any security mechanism, steal confidential information, install malware, and even disable the entire system. In this lecture, security researcher Benny Seltzer from Intel will present a study presented at the international DEFCON conference, we will talk about how they discovered security problems that left billions of devices exposed to danger, about the journey to crack the defenses, to run code in SMM mode, and we will see the successes as well as the unsuccessful attempts along the way. At the end, we will explain how the attack can be carried out with the privileges of an average user by chaining several additional vulnerabilities together.\nSunday 07/28 at 18:30 in the piano auditorium,&nbsp;\nTaub building. For details and registration: https://forms.gle/iGJw6Nwae7s7sWHu7\nEverybody is invited!&nbsp;\nThe lecture is also suitable for beginners without background in the field
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub, 0 floor
UID:eventx6a5a287eec27210654
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20240730T100000
DTEND;TZID=Asia/Jerusalem:20240730T110000
DTSTAMP;TZID=Asia/Jerusalem:20240730T100000
SUMMARY: MSC  talk by Yair Davidson  On the Hölder Stability of Multiset and Graph Neural Networks  at 2024-07-30 10:00:00
DESCRIPTION:Famously, multiset neural networks based on sum-pooling can separate all distinct multisets, and as a result can be used by message passing neural networks (MPNNs) to separate all pairs of graphs that can be separated by the 1-WL graph isomorphism test.&nbsp;However, the quality of this separation may be very weak, to the extent that the embeddings of "separable" multisets and graphs might even be considered identical when using fixed finite precision.\nIn this work, we propose to fully analyze the separation quality of multiset models and MPNNs via a novel adaptation of Lipschitz and H&ouml;lder continuity to parametric functions.&nbsp;We prove that common sum-based models are lower-H&ouml;lder continuous, with a H&ouml;lder exponent that decays rapidly with the network's depth.&nbsp;Our analysis leads to adversarial examples of graphs which can be separated by three 1-WL iterations, but cannot be separated in practice by standard maximally powerful MPNNs.&nbsp;To remedy this, we propose two novel MPNNs with improved separation quality, one of which is lower Lipschitz continuous.&nbsp;We show these MPNNs can easily classify our adversarial examples, and compare favorably with standard MPNNs on standard graph learning tasks.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 8, https://technion.zoom.us/j/97109293645
UID:eventx6a5a287eec27d10645
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20240730T143000
DTEND;TZID=Asia/Jerusalem:20240730T153000
DTSTAMP;TZID=Asia/Jerusalem:20240730T143000
SUMMARY: colloq  talk by Roy Schwartz (HUJI)  Green AI  at 2024-07-30 14:30:00
DESCRIPTION:The computations required for deep learning research have been doubling every few months, resulting in an estimated 5,000x increase from 2018 to 2022. This trend has led to unprecedented success in a range of&nbsp;AI&nbsp;tasks. In this talk I will discuss a few troubling side-effects of this trend, touching on issues of lack of inclusiveness within the research community, and an increasingly large environmental footprint. I will then present&nbsp;Green&nbsp;AI&nbsp;&ndash; an alternative approach to help mitigate these concerns.&nbsp;Green&nbsp;AI&nbsp;is composed of two main ideas: increased reporting of computational budgets, and making efficiency an evaluation criterion for research alongside accuracy and related measures. I will focus on the latter topic, discussing some of our recent efforts for reducing the costs of&nbsp;AI. This is joint work with Michael Hassid, Yossi Adi, Matanel Oren, Tal Remez, Jesse Dodge, Noah A. Smith, Oren Etzioni and Jonas Gehring.&nbsp;\nBio: Roy&nbsp;Schwartz&nbsp;is a senior lecturer (assistant professor) at the School of Computer Science and Engineering at The Hebrew University of Jerusalem (HUJI).&nbsp;Roy&nbsp;studies natural language processing and&nbsp;artificial&nbsp;intelligence. Prior to joining HUJI,&nbsp;Roy&nbsp;was a postdoc (2016-2019) and then a research scientist (2019-2020) at the Allen institute for&nbsp;AI&nbsp;and at The University of Washington, where he worked with Noah A. Smith.&nbsp;Roy&nbsp;completed his Ph.D. in 2016 at HUJI, where he worked with Ari Rappoport.&nbsp;Roy&rsquo;s work has appeared on the cover of the CACM magazine, and has been featured, among others, in the New York&nbsp;Times, Haaretz, and Ynet.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 337
UID:eventx6a5a287eec28910629
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20240802T100000
DTEND;TZID=Asia/Jerusalem:20240802T120000
DTSTAMP;TZID=Asia/Jerusalem:20240802T100000
SUMMARY: CSpecial Event  Orientation day of the Faculty of Computer Science, Technion  at 2024-08-02 10:00:00
DESCRIPTION:Do you know someone who wants to do something big?\nOrientation day for a bachelor's degree in the Faculty of Computer Science at the Technion is approaching and you are all invited!\nFriday 2.8 | Sharona, Tel Aviv \nFor more details and to register: https://lp.technion.ac.il/computer_science
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Azrieli Sarona, Tel Aviv
UID:eventx6a5a287eec29610653
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20240805T153000
DTEND;TZID=Asia/Jerusalem:20240805T163000
DTSTAMP;TZID=Asia/Jerusalem:20240805T153000
SUMMARY: pixel-club  talk by Yoel Shkolnisky (Tel Aviv University)  Object detection under the linear subspace model  at 2024-08-05 15:30:00
DESCRIPTION:Detecting unknown objects in noisy data is a key task in many problems. A natural model for the unknown objects is the linear subspace model, which assumes that the objects can be expanded in some known basis (such as the Fourier basis). In this talk, I will present an object detection algorithm that under the linear subspace model is asymptotically guaranteed to find all objects while making only a small percentage of false discoveries. We demonstrate our derivations for the problem of particle picking in cryo-electron microscopy.\nLight refreshments will be served at 15:00 at the EE faculty lounge on the 8th floor.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Amado 232
UID:eventx6a5a287eec2a110662
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20240806T100000
DTEND;TZID=Asia/Jerusalem:20240806T110000
DTSTAMP;TZID=Asia/Jerusalem:20240806T100000
SUMMARY: PHD  talk by Ori Roth  Type Automata  at 2024-08-06 10:00:00
DESCRIPTION:Modern programming languages rely on advanced and intricate type systems.Expressive type systems support more language features but often result in unexpected language capabilities.&nbsp;For example, we know that the Java type system is Turing complete, which means it is so complex that Java compilers cannot guarantee termination.&nbsp;But a powerful type system can also be a blessing in disguise.&nbsp;Over the years, crafty programmers found ways to coerce type systems to perform basic computations at the type level.&nbsp;Certain program interfaces employ this form of metaprogramming to detect and report specialized program bugs before runtime.\nIn classical computer science, the notion of computational power is tightly coupled with automata (abstract machines) such as finite-state and Turing machines.&nbsp;We reason about the computability of various type systems by connecting them with well-founded classes of automata through type automata -- machines that employ program types as control and storage.&nbsp;By describing a bisimulation between type automata and classical automata, we precisely classify the expressiveness of a type system.\nThis work is comprised of a series of studies that analyze different type systems using type automata.&nbsp;We classify the computability of the decidable type systems fragments of Kennedy and Pierce in terms of regular and context-free tree languages.&nbsp;On the other hand, we prove that the Python type system defined in PEP 484 is Turing complete.&nbsp;In addition, we introduce novel metaprogramming techniques for an advanced interface design called fluent API.&nbsp;Our fluent APIs can enforce the API protocol or the grammar of an embedded domain-specific language (DSL) at compile time.&nbsp;We present the first fluent API design that supports all deterministic context-free API protocols and DSLs.&nbsp;We also demonstrate how to create elegant and sophisticated fluent APIs in functional programming languages.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 8
UID:eventx6a5a287eec2ab10655
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20240806T110000
DTEND;TZID=Asia/Jerusalem:20240806T120000
DTSTAMP;TZID=Asia/Jerusalem:20240806T110000
SUMMARY: MSC  talk by Shahar Mor  Statistically dense intervals in binary sequences with applications to assessing local enrichment in the human genome  at 2024-08-06 11:00:00
DESCRIPTION:Statistical enrichment tools are highly useful in biological research. Current approaches to statistical enrichment in ranked or ordered lists such as, for example, GSEA and GOrilla, are limited to the suffix (prefix) of the list. These methods assess extreme density of 1s in binary vectors on either side. Statistical significance can be assigned using, e.g, Wilcoxon Rank Sum and mHG statistics.In this work we extend the mHG approach to also address enrichment in any index intervals of the binary vector. We define and provide a partial characterization of related distributions under a uniform null model. Our partial characterization yields useful bounds for extreme events. We provide a software tool to the community, implementing the method in Python. Finally, we analyze several example use cases and describe the results. We show, for example, that lung cancer differential expression, comparing ADC to other types, is enriched in a region of Chromosome 3. This example represents a typical use case for imHG -- obtaining enriched intervals for any set of genes of interest. We provide a Python implementation, called imHG, for finding and reporting enriched genomic intervals with any given list of genes of interest.&nbsp;\n&nbsp;
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:https://technion.zoom.us/my/s4057666386
UID:eventx6a5a287eec2b810641
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20240806T183000
DTEND;TZID=Asia/Jerusalem:20240806T210000
DTSTAMP;TZID=Asia/Jerusalem:20240806T183000
SUMMARY: CSpecial Event  End Of Year Festival  at 2024-08-06 18:30:00
DESCRIPTION:The Computer Science Student Council brings you the "The Good, the Bad and the Ugly" festivalTuesday, August 66:30 p.m. Lecturers pour beers, free food and drink, karaoke and surprises (admission for students from MDAMH only), on the Taub terrace.9:00 p.m. party in the transparent hall (with ticket presentation only)\nThe event is sponsored by Plus500
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub balcony + the transparent hall in the student house
UID:eventx6a5a287eec2c410664
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20240807T113000
DTEND;TZID=Asia/Jerusalem:20240807T123000
DTSTAMP;TZID=Asia/Jerusalem:20240807T113000
SUMMARY: ceClub  talk by Shir Landau Feibish (The Open University)  Time-Aware Network Telemetry in the Data Plane  at 2024-08-07 11:30:00
DESCRIPTION:Collecting network telemetry&nbsp;is essential for detecting problems in the network. In recent years we have seen an abundance of research on network telemetry in the data plane. Many of these solutions analyze traffic continuously over a long period of time, while resetting&nbsp;the structure from time to time. If shorter time intervals are needed, sliding windows are usually used, yet these incur significant resource and management overhead. However, often in order to understand what is happening&nbsp;in the network we need to measure events that have recently&nbsp;happened.&nbsp;\nWe explore the concept of time-aware network telemetry and propose several different paradigms that we utilize for various telemetry tasks. The first is a reactive monitoring scheme for packet loss detection, that monitors only when needed, by tailoring the measurement interval to when losses may occur. The second is a mechanism for finding heavy hitter flows of the recent&nbsp;past. Finally, we explore the option of deleting and decaying a set-membership data structure to allow for more updated information to be maintained.&nbsp;\nShir is a Senior Lecturer (Assistant Professor) and head of the RUNS&nbsp;lab at The Open University of Israel. Before joining the Open University Shir was a postdoctoral researcher at Princeton University, where she worked with Prof. Jennifer Rexford. She received her Ph.D. from Tel Aviv University, where she was advised by Prof. Yehuda Afek in collaboration with Prof. Anat Bremler-Barr.\nShir's main interests are network monitoring and management with a focus on programmable networks.Shir has received several awards including the Eric and Wendy Schmidt Postdoctoral Award for Women in Mathematical and Computing Sciences and was named one of the Rising Stars in Networking and Communications of 2020 by the N2Women Organization.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zisapel 506
UID:eventx6a5a287eec2d610658
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20240808T133000
DTEND;TZID=Asia/Jerusalem:20240808T143000
DTSTAMP;TZID=Asia/Jerusalem:20240808T133000
SUMMARY: MSC  talk by Or Keret  Doubly-Efficient Batch Verification in Statistical Zero-Knowledge  at 2024-08-08 13:30:00
DESCRIPTION:A sequence of recent works, concluding with Mu et al. (Eurocrypt, 2024) has shown that every problem $\Pi$ admitting a non-interactive statistical zero-knowledge proof (NISZK) has an efficient zero-knowledge \emph{batch verification} protocol. Namely, an NISZK protocol for proving that $x_1, \dots, x_k \in \Pi$ with communication that only scales poly-logarithmically with $k$. A caveat of this line of work is that the prover runs in exponential-time, whereas for NP problems it is natural to hope to obtain a \emph{doubly-efficient proof} -- that is, a prover that runs in polynomial-time given the $k$ NP witnesses.\nIn this work we show that every problem in NISZK $\cap$ UP has a \emph{doubly-efficient} interactive statistical zero-knowledge proof with communication $\poly(n, \log(k))$ and $\poly(\log(k), \log(n))$ rounds. The prover runs in time $\poly(n, k)$ given access to the $k$ UP witnesses. Here $n$ denotes the length of each individual input, and UP is the subclass of NP relations in which YES instances have unique witnesses.\nThis result yields doubly-efficient statistical zero-knowledge batch verification protocols for a variety of concrete and central cryptographic problems from the literature.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:https://technion.zoom.us/j/97383546152
UID:eventx6a5a287eec2e810656
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20240812T123000
DTEND;TZID=Asia/Jerusalem:20240812T133000
DTSTAMP;TZID=Asia/Jerusalem:20240812T123000
SUMMARY: PHD  talk by Konstantin Zabarnyi  Information Design in the 21st Century  at 2024-08-12 12:30:00
DESCRIPTION:The field of information design studies strategic information revelation by a certain sender to receivers. The Bayesian persuasion model, introduced by Kamenica and Gentzkow, assumes that the sender can trustworthily commit to a randomized information revelation policy, called a signaling scheme. In contrast, the cheap talk model assumes that the sender does not have such a commitment power. The research focuses on optimizing the sender&rsquo;s utility.In this bundle of works, we study versions of the Bayesian persuasion and the cheap talk model motivated by various challenges of the 21st century. For a single-receiver cheap talk setting, we show that approximating the sender-optimal equilibrium utility up to a certain additive or multiplicative constant is NP-hard. It comes in a sharp contrast to the analogous well-studied problem in Bayesian persuasion, which is computationally tractable. Since the commitment power is unrealistic in some real-life scenarios, this result might seem to limit the extent of applicability of information design. Luckily, we manage to get positive computational results in some natural special cases.For a single-receiver Bayesian persuasion setting, we provide a simple explicit formula for the sender&rsquo;s minimal regret when the receiver&rsquo;s utility function is chosen adversarially. We further provide positive computational results when the sender&rsquo;s choice of the signaling scheme is limited by privacy constraints. For a multi-receiver Bayesian persuasion model, we consider a setting in which the sender can communicate with the receivers via several (possibly overlapping) communication channels; such a setting can be motivated by social media interactions. We show that in general, the sender&rsquo;s optimization problem in this multi-channel setting is harder than both private and public Bayesian persuasion. Still, finding a sender&rsquo;s optimal signaling scheme is tractable for several special cases corresponding to communication in hierarchical organizations.Finally, we combine information design with two further burning topics: contract design and information aggregation. In the principal-agent setting in contract design, we provide a full characterization of all the implementable utility profiles and agent&rsquo;s actions when the information revealed to the principal about the agent&rsquo;s action is chosen by a social planner. In information aggregation, we show that the robustly optimal way to aggregate anonymous binary recommendations from symmetric agents with adversarially-correlated private information about a hidden state is the random dictator rule: picking one recommendation uniformly at random and following it.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 8 &amp; https://technion.zoom.us/j/95520720819&nbsp;
UID:eventx6a5a287eec30910663
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20240814T190000
DTEND;TZID=Asia/Jerusalem:20240814T220000
DTSTAMP;TZID=Asia/Jerusalem:20240814T190000
SUMMARY: CSpecial Event  Band Night of the Faculty of Computer Science  at 2024-08-14 19:00:00
DESCRIPTION:Invitation to Band Night\nYou are invited to Band Night of the Faculty of Computer Science!!Talented students and faculty members from our department will perform in a variety of musical ensembles and styles.\nJoin us for an unforgettable evening of live performances and amazing energy.\nWednesday, August 14th at 7:00 PM on the Taub Terrace.\nThere are sheltered areas, excellent air quality, so get your strings ready &ndash; entrance is free!\nBest regards,The Student Committee
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub Terrace 2nd floor
UID:eventx6a5a287eec31d10666
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20240815T105000
DTEND;TZID=Asia/Jerusalem:20240815T200500
DTSTAMP;TZID=Asia/Jerusalem:20240815T105000
SUMMARY: CSpecial Event  Registration for the 2024 Excellent Project competition is underway! The submission deadline has been extended!  at 2024-08-15 10:50:00
DESCRIPTION:Did you do an innovative, interesting, groundbreaking project?\nYou are invited to apply at the link: https://tinyurl.com/cs-projects24\nApply by August 15, 2024\nThe final phase will take place on August 21, 2024 - noon on Wednesday, in the format of a project fair - the participation of the contestants is mandatory. The first place winners will receive a cash prize from the Amdocs company.\nGood luck to all participants!
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub
UID:eventx6a5a287eec32910644
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20240820T113000
DTEND;TZID=Asia/Jerusalem:20240820T123000
DTSTAMP;TZID=Asia/Jerusalem:20240820T113000
SUMMARY: pixel-club  talk by Ben Finkelshtein (University of Oxford)  Cooperative Graph Neural Networks  at 2024-08-20 11:30:00
DESCRIPTION:Graph neural networks are popular architectures for graph machine learning, based on iterative computation of node representations of an input graph through a series of invariant transformations. A large class of graph neural networks follow a standard message-passing paradigm: at every layer, each node state is updated based on an aggregate of messages from its neighborhood. In this work, we propose a novel framework for training graph neural networks, where every node is viewed as a player that can choose to either &lsquo;listen&rsquo;, &lsquo;broadcast&rsquo;, &lsquo;listen and broadcast&rsquo;, or to &lsquo;isolate&rsquo;. The standard message propagation scheme can then be viewed as a special case of this framework where every node &lsquo;listens and broadcasts&rsquo; to all neighbors. Our approach offers a more flexible and dynamic message-passing paradigm, where each node can determine its own strategy based on their state, effectively exploring the graph topology while learning. We provide a theoretical analysis of the new message-passing scheme which is further supported by an extensive empirical analysis on synthetic and real-world data.\nUnder the supervision of Prof. Michael Bronstein and Dr. Ismail Ilkan Ceylan.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:1061, Meyer Building
UID:eventx6a5a287eec33410669
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20240821T103000
DTEND;TZID=Asia/Jerusalem:20240821T113000
DTSTAMP;TZID=Asia/Jerusalem:20240821T103000
SUMMARY: MSC  talk by Sari Hleihil  First Priniciple Based Geometric Deep Learning  at 2024-08-21 10:30:00
DESCRIPTION:Geometric Deep Learning attempts to apply deel learning methodlogies to domains where a grid structure doesn't exit. We advocate for using principled methods to define the primitives of these networks. As such we define networks that stem from the symmetries of geometric representations, And show how analyzing some of these primitives spectrally reveals that combining allows for SOTA performance. This is finally complemented by the introduction of a suite of identity losses that are customizable and induced from geometric quantities.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:https://technion.zoom.us/j/91974380111&nbsp; &amp; Taub 9
UID:eventx6a5a287eec33f10668
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20240821T123000
DTEND;TZID=Asia/Jerusalem:20240821T143000
DTSTAMP;TZID=Asia/Jerusalem:20240821T123000
SUMMARY: CSpecial Event  The annual projects fair of the Faculty of Computer Science and the Outstanding Project competition  at 2024-08-21 12:30:00
DESCRIPTION:Hello everyone,\nWe are pleased to invite you to the annual projects fair of the Taub Faculty of Computer Science, along with the Outstanding Project competition&mdash;Wednesday, August 21, starting at 12:30 PM in the Taub Lobby - Floor 0.\nEveryone is welcome to support the competing teams and be impressed by significant and creative projects.\nWe look forward to seeing you!
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub lobby, floor 0
UID:eventx6a5a287eec34910667
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20240821T130000
DTEND;TZID=Asia/Jerusalem:20240821T140000
DTSTAMP;TZID=Asia/Jerusalem:20240821T130000
SUMMARY: TDC Seminar  talk by Tal Yankovitz (Tel Aviv University)  A stronger bound for linear 3-LCC  at 2024-08-21 13:00:00
DESCRIPTION:A q-locally correctable code (LCC) C:{0,1}^k-&gt;{0,1}^n &nbsp;is a code in which it is possible to correct every bit of a (not too) corrupted codeword by making at most q queries to the word. The cases in which q is constant are of special interest, and so are the cases that C is linear.\nIn a breakthrough result Kothari and Manohar (STOC 2024) showed that for linear 3-LCC n=2^&Omega;(k^1/8) . In this work we prove that n=2^&Omega;(k^1/4) . As Reed-Muller codes yield 3-LCC with n=2^O(k^1/2) , this brings us closer to closing the gap. Moreover, in the special case of design-LCC (into which Reed-Muller fall) the bound we get is n=2^&Omega;(k^1/3).
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Amado 814
UID:eventx6a5a287eec35310661
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20240821T173000
DTEND;TZID=Asia/Jerusalem:20240821T183000
DTSTAMP;TZID=Asia/Jerusalem:20240821T173000
SUMMARY: PHD  talk by Igor Smolyar  Improving the Performance of Direct Memory Accesses, Interrupts, and Paravirtualization of High-Performance I/O  at 2024-08-21 17:30:00
DESCRIPTION:The rapid growth of data-intensive applications and high-speed I/O devices has led to increasing demands on I/O performance in both virtualized cloud environments and bare metal setups. But existing systems struggle to fully exploit the potential of modern hardware due to inefficiencies at various I/O stack layers. This thesis presents three novel techniques that optimize I/O performance across virtualized and bare metal environments: IOctopus, cinterrupts, and Hermes.\nIOctopus eliminates non-uniform DMA (NUDMA) effects in multi-CPU systems by connecting I/O devices to every CPU and abstracting multiple PCIe endpoints into a single entity. IOctopus eliminates all remote DMAs, transforming them to local operations, thus improving I/O throughput and latency by as much as 2.7x and 1.28x, respectively.\nCinterrupts enable fine-grained control over interrupt generation in modern high-speed storage devices by allowing software to indicate which I/O requests are latency sensitive. With this information, the device can &ldquo;calibrate&rdquo; its interrupts to completions of latency-sensitive operations. This approach increases throughput, reduces CPU consumption, and achieves lower latency even when interrupts are coalesced. While primarily designed for NVMe SSDs, the cinterrupts principle can be extended to other I/O technologies. Calibrated interrupts increase throughput by up to 35%, reduce CPU consumption by as much as 30%, and achieve up to 37% lower latency even when interrupts are coalesced.\nIn high-throughput virtual setups, operators may choose to employ dedicated hypervisor cores &ndash; denoted &ldquo;sidecores&rdquo; &ndash; in order to process the network I/O of virtual machines (VMs). Sidecores reduce virtualization overheads by eliminating architectural exits (and instead polling on virtual queues),\nand by reducing the number of context switches between virtual CPUs and their corresponding virtual I/O processing threads. The problem is that the decision of whether or not to use sidecores, as well as their number, is determined statically in existing systems, thereby significantly limiting the applicability of this optimization. We solve this problem by introducing Hermes, which adapts to changing workloads and matches the performance of optimally-tuned static configurations at any point in time. Hermes improves throughput by as much as 12x, reduces CPU consumption by up to 20%, and shortens tail latency by at most 63%.\nTogether, IOctopus, cinterrupts, and Hermes represent an advancement in I/O optimization, enabling both cloud providers and bare metal operators to better utilize modern high-speed I/O devices, including advanced storage systems, and improve overall system performance.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:https://technion.zoom.us/j/96726470488
UID:eventx6a5a287eec35e10665
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20240823T113000
DTEND;TZID=Asia/Jerusalem:20240823T123000
DTSTAMP;TZID=Asia/Jerusalem:20240823T113000
SUMMARY: pixel-club  talk by Rotem Benisty  On Feature Extraction from MRI Data of Crohn’s Disease patients  at 2024-08-23 11:30:00
DESCRIPTION:Diagnosing Crohn's Disease (CD) typically involves examining 2D slices from magnetic resonance enterography (MRE). However, the anisotropic resolution of MRE complicates precise 3D measurements and visualization. The absence of automated 3D measurement systems further complicates assessment. Previous methods for generating isotropic volumes from anisotropic data often rely on extensive 3D data and focus solely on interslice resolution, leading to suboptimal outcomes due to data scarcity and inaccuracies. We propose a self-supervised multi-plane generative model using Generative Adversarial Network (GAN) architecture, incorporating multiple discriminators for different planes. We introduce a semi-automatic algorithm to predict the centerline of the terminal ileum, which is the part of the body primarily affected by CD, enhancing efficiency in 3D MRE analysis. Evaluation using 115 2D abdominal MRE datasets from Rambam Health Care Campus underscores the potential of our approach to enhance diagnostic accuracy and visualization in CD. Our semi-automatic centerline prediction reduces the radiologist's analysis time significantly. Additionally, our isotropic volume generation model can be expanded to other anatomical regions, thereby reducing MRI scan-time and costs.&nbsp;
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Link
UID:eventx6a5a287eec36d10646
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20240827T110000
DTEND;TZID=Asia/Jerusalem:20240827T120000
DTSTAMP;TZID=Asia/Jerusalem:20240827T110000
SUMMARY: PHD  talk by Ben Galili  Stable Statistics, DNA, Feature Selection, and the in-between  at 2024-08-27 11:00:00
DESCRIPTION:In this talk, I will present topics from my PhD work. The talk consists of three topics:Feature selection is a core process in building machine learning models. It is often essential for optimizing performance and sometimes essential to support practical use, such as when the number of features to be measured in the execution stage is limited by hardware or other factors. We examined the procedure in challenging situations: feature selection on a very high-dimensional dataset, feature selection in a privacy-preserving computational environment, and feature selection for a normalization-free classifier.The second part of the work focuses on stable statistics. Statistically significant results supporting the rejection of the null hypothesis are achieved when the probability of obtaining the test results, or more extreme results, is unlikely to have occurred under the null hypothesis. This probability is the p-value of our observation. Inferred p-values, however, can be very sensitive to inaccuracies in the input data. We introduced and defined the uncertainty that arises from sample labeling errors in the context of statistical tests in a deterministic and stochastic approach. We developed algorithms for efficiently calculating a stability interval around the original p-value. We also developed an algorithm for selecting a differential expression gene set that is robust to label errors.In the last part of this work, we applied statistical and computational methods in computational biology&mdash;virus classifier, aspects of composite DNA synthesis, and a new method for finding gene expression enrichment.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:https://runi-ac-il.zoom.us/j/86065739117
UID:eventx6a5a287eec37910657
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20240827T130000
DTEND;TZID=Asia/Jerusalem:20240827T140000
DTSTAMP;TZID=Asia/Jerusalem:20240827T130000
SUMMARY: MSC  talk by Oriel Limor  An Interpretation of Spearman Correlation via k-Subset Permutations and The Edge Collector’s Problem  at 2024-08-27 13:00:00
DESCRIPTION:This talk will cover two topics:First, we suggest a new interpretation of Spearman correlation using k-subset permutations. We characterize the distribution of the Spearman correlation of a permutation that starts with the identity permutation over n elements and is perturbed by uniformly shuffling a random subset of k indices. We present some key aspects of this distribution, including its expected value and variance for every k between 1 and n. We then generalize it to any starting vector.In the second part of the talk, we define a generalization of the &lsquo;Coupon Collector&rsquo; problem dubbed &lsquo;Edge Collector&rsquo;, where we are given two sets A and B of the same size, and a perfect matching between them. At each step we draw a pair of elements from A, and get their pair of matching elements in B. However, we do not know which element from the first pair corresponds to which element in the second. Using a Markov process, we find a recursive formula and an estimation for the expected number of steps needed to determine all the matchings.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 8
UID:eventx6a5a287eec38510672
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20240828T153000
DTEND;TZID=Asia/Jerusalem:20240828T163000
DTSTAMP;TZID=Asia/Jerusalem:20240828T153000
SUMMARY: PHD  talk by Eytan Singher  Effective Automatic Deductive Reasoning Using Equality Saturation  at 2024-08-28 15:30:00
DESCRIPTION:Automated deductive reasoning plays an important role in software verification, optimization, and mathematical theorem discovery. This talk explores novel applications and extensions of equality saturation, a technique for efficiently representing and manipulating large sets of equivalent expressions, to enhance automatic deductive reasoning across various domains.In this talk, we present a collection of three works:\nA symbolic theory exploration system, dubbed Thesy, based on equality saturation that efficiently discovers and proves new mathematical lemmas.By leveraging equality saturation to manipulate symbolic values, TheSy outperforms testing-based approaches in both speed and scope.The second work is a novel extension to e-graphs, the data structure at the base of the equality saturation technique, enabling efficient conditional reasoning within the equality saturation framework. This memory-efficient approach to representing multiple e-graphs simultaneously opens new avenues for applying equality saturation to complex conditional reasoning problems.Lastly, we present the Lightweight Equality Saturation (LES) prover. LES utilizes the immense amount of lemmas and definitions available in proof assistant environments, as well as some additional properties resulting from the high-order logic used in these environments to go toe-to-toe with state-of-the-art provers.\nThese innovations synergize to create a more comprehensive approach to deductive reasoning, from expanding underlying theories to handling conditional logic and enhancing theorem proving capabilities in interactive environments.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:https://technion.zoom.us/j/96493253890?pwd=WXHqRoJ2A2ssoyWYUR9AjNEePyl4HH.1 &amp; Taub 9
UID:eventx6a5a287eec39010670
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20240901T113000
DTEND;TZID=Asia/Jerusalem:20240901T123000
DTSTAMP;TZID=Asia/Jerusalem:20240901T113000
SUMMARY: PHD  talk by Thomas Dages  Unleashing Asymmetry unto Metric-based Geometric Deep Learning  at 2024-09-01 11:30:00
DESCRIPTION:Metric theory offers powerful tools for analysing and processing shapes on curved manifolds, with Riemannian metrics being the most commonly used due to their simplicity and success in many applications. However, Riemannian metrics are limited by their symmetric nature, where lengths of paths are independent of traversal direction. Finsler metrics, a generalisation including asymmetric distances, provide broader tools but are rarely applied in practice, perhaps due to their theoretical complexity. This talk describes Finsler metrics, showcasing their potential through two applications. First, we revisit anisotropy in Laplace-Beltrami operators (LBO) on Riemannian manifolds. By exploring a Finsler perspective, we design a new Riemannian LBO with Finsler-movitaved anisotropy, and apply it for shape matching. Second, we revisit heuristic deformation strategies in image convolution, proposing a unifying theory where kernel positions are samples of unit balls from implicit metrics. Introducing metric convolutions, which involve sampling of unit balls of explicit metrics and are compatible with neural networks, we achieve competitive results in denoising and classification tasks.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 9
UID:eventx6a5a287eec39c10671
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20240902T143000
DTEND;TZID=Asia/Jerusalem:20240902T153000
DTSTAMP;TZID=Asia/Jerusalem:20240902T143000
SUMMARY: MSC  talk by Tamir Shor  Learning In Joint Input And Downstream Task Optimization  at 2024-09-02 14:30:00
DESCRIPTION:Traditional discriminative Machine-Learning approaches often formulate problems as optimization of a parameterized function mapping from a given input space to some desired output space. While this formulation is applicable to many theoretical and practical problems, it is inherently reliant on the assumption that the dataset X is constant. The world around us is abundant with scenarios where this is either not the case, or where dropping this assumption could help achieve better solutions.Namely, in many scenarios one can gain added benefit from starting from one step before the data is acquired - if the way data is perceived by our sensors can be modeled, with a correct model it can be parameterized. If it can be parameterized, with the right mathematical tools (e.g. Deep Learning), it can be optimized to gives us datasets more useful for achieving the downstream task athand. As we are solving task-specific different optimization of the acquired data, a joint optimization of the downstream task model and data acquisition model is usually called for. This type of joint optimization calls for the delicate balancing in optimization of the two systems (acquisition and downstream), bringing upon a rich and interesting range of problems, coming from a diverse set of fields in science and engineering.In our research we identify and study a collection of such problems, and offer modeling and optimization schemes to solve them.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:https://technion.zoom.us/j/95866375556&nbsp;\n&nbsp;
UID:eventx6a5a287eec3a710639
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20240904T103000
DTEND;TZID=Asia/Jerusalem:20240904T113000
DTSTAMP;TZID=Asia/Jerusalem:20240904T103000
SUMMARY: pixel-club  talk by Tal Daniel  Variational Generative Approaches to Self-Supervised Representation Learning: From Introspective Training to Object-Centric Learning  at 2024-09-04 10:30:00
DESCRIPTION:Unsupervised latent variable models serve as highly effective tools for representing complex data such as images or videos, relevant for applications such as robotic manipulation, video generation, novelty detection, and many more. Variational Autoencoders (VAEs) provide compact latent representations with stability and efficiency. In this talk, we will explore modern VAEs that mitigate shortcomings of classical approaches such as blurry images, and can be used as a basis for strong world models. The first paper (CVPR 2021 Oral) introduces &ldquo;Soft-IntroVAE&rdquo; , a refined approach to introspective variational autoencoders, enhancing training stability and theoretical insights while showcasing its applications. The second paper (ICML 2022) presents &ldquo;Deep Latent Particles (DLP)&rdquo; for unsupervised image representation learning, a new VAE where the latent space is keypoint-based, offering disentangled object features, uncertainty estimation, and versatile applications. Building on DLP, the third paper (TMLR 2024) presents &ldquo;DDLP&rdquo;, a novel extension to video prediction, manipulation and generation, where a differentiable tracking module is employed over particles to drive the dynamics modeling. We conclude by highlighting recent applications of these representations in deep reinforcement learning and biomedical domains, demonstrating their broad impact and potential for future research.\nTal (https://taldatech.github.io) is a final-year Ph.D. student in the Electrical and Computer Engineering faculty at the Technion, where he earned his B.Sc. and M.Sc., under the supervision of Prof. Aviv Tamar. He is the winner of The Miriam and Aaron Gutwirth Memorial and The Irwin and Joan Jacobs Ph.D fellowships. His research interests include unsupervised representation learning, generative modeling and reinforcement learning.\nPh.D. Under the supervision of Prof. Aviv Tamar.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:506, Zisapel Building &amp; Zoom Link
UID:eventx6a5a287eec3b410675
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20240904T113000
DTEND;TZID=Asia/Jerusalem:20240904T123000
DTSTAMP;TZID=Asia/Jerusalem:20240904T113000
SUMMARY: ceClub  talk by Lior Zeno  Enhancing Distributed Systems with In-Network Computing  at 2024-09-04 11:30:00
DESCRIPTION:The emerging paradigm of in-network computing leverages programmable network hardware, such as switches, to enhance the performance, reliability, and functionality of distributed systems. Traditionally, networks in distributed systems have been treated merely as conduits for data, with limited assumptions about their capabilities. However, by introducing higher-level abstractions that utilize the potential of programmable network devices, significant improvements in distributed applications can be achieved.This presentation will cover three key systems: SwiSh, SwitchV2P, and SwitchBFT. SwiSh provides a distributed state management layer within programmable switches, delivering substantial gains in update throughput and latency. SwitchV2P optimizes virtual network performance through in-network caching for address translation, significantly reducing flow completion times and network overhead. Finally, SwitchBFT demonstrates how trusted network switches can enable Byzantine Fault Tolerant consensus with the speed and efficiency of Crash Fault Tolerance protocols.This work highlights the transformative potential of in-network computing in advancing distributed applications and data center infrastructure.Lior Zeno is a PhD student under the supervision of Prof. Mark Silberstein.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:https://technion.zoom.us/j/94673013539 &amp; Zisapel 506
UID:eventx6a5a287eec3c110674
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20240909T180000
DTEND;TZID=Asia/Jerusalem:20240909T200000
DTSTAMP;TZID=Asia/Jerusalem:20240909T180000
SUMMARY: CSpecial Event  Information Session for Students who interested in Advanced Degrees in AI and Machine Learning  at 2024-09-09 18:00:00
DESCRIPTION:Information Session for students who interested in advanced degrees in Artificial Intelligence and Machine Learning\nWhether you're contemplating how to enter the field and unsure where to start, or if you've already completed a Master's degree and are considering starting a PhD, we invite you to join us!\nDate: Monday, September 9th at 6:00 PMProgram: Scientific lectures, student panel, and mingling.\nRegistration: https://forms.gle/LwHYxj6izzifbhxh8
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub
UID:eventx6a5a287eec3cd10676
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20240911T113000
DTEND;TZID=Asia/Jerusalem:20240911T123000
DTSTAMP;TZID=Asia/Jerusalem:20240911T113000
SUMMARY: ceClub  talk by Anan Kabaha  Verification of Robustness Properties for Neural Networks  at 2024-09-11 11:30:00
DESCRIPTION:Deep neural networks are successful in various tasks but are also susceptible to adversarial examples: malicious input perturbations designed to deceive the network. Many adversarial attacks on image classifiers involve making imperceptible changes bounded by a small &epsilon; with respect to an Lₚ&nbsp;norm (e.g., p = 0, 1, 2, &infin;), by a small interval neighborhood, or by semantic feature perturbations, such as adjustments in brightness, translation, or rotation. To understand the robustness of a DNN to adversarial examples, most existing works propose to analyze the network's local robustness of a given &epsilon;-ball. Despite the significant progress in their efficiency and precision, existing verifiers of &epsilon;-ball neighborhoods are limited to verifying small neighborhoods and most of them do not provide global guarantees. In our thesis, we propose several verifiers of different kinds of robustness properties for neural networks: a verifier for computing maximally locally robust interval neighborhoods, a verifier for computing maximally locally robust feature neighborhoods, a verifier for computing global robustness guarantees to different kinds of perturbations, a system relying on global robustness verification to protect the privacy of neural networks' training sets, and a system relying on global robustness verification to provide formal guarantees over the reliability of neural network quantization schemes.\nAnan is a PhD student supervised by Prof. Dana Drachsler Cohen.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zisapel 506; andhttps://us06web.zoom.us/j/5959644772?pwd=LbJTHb0WfPzI9Zdn6q4d0Se9I69j3g.1&nbsp;
UID:eventx6a5a287eec3d810677
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20240915T160000
DTEND;TZID=Asia/Jerusalem:20240915T170000
DTSTAMP;TZID=Asia/Jerusalem:20240915T160000
SUMMARY: MSC  talk by Liav Zafar  One-Message Secure Reductions  at 2024-09-15 16:00:00
DESCRIPTION:How can two parties jointly sample from a source of correlated randomness (X,Y), say two hands of cards in a poker game, without leaking any extra information? This secure sampling question is strongly motivated by the design of efficient protocols for secure computation, allowing the two parties to compute a function of their secret inputs without revealing their inputs to each other.While there has been major progress on securely sampling some useful correlations, others seem much more costly to generate. This motivates the study of efficient techniques for securely converting copies of a given source correlation into copies of a given target correlation.\nThis talk will discuss a systematic study of such secure conversion protocols that involve only a single message. We present a general rejection-sampling based technique for designing such protocols, and apply them towards improving the communication complexity of distributed symmetric cryptography. On the negative side, we show lower bounds on the communication complexity for such conversions, matching our positive results up to small constant factors.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:https://technion.zoom.us/j/5655724639
UID:eventx6a5a287eec3e410678
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20240918T113000
DTEND;TZID=Asia/Jerusalem:20240918T123000
DTSTAMP;TZID=Asia/Jerusalem:20240918T113000
SUMMARY: pixel-club  talk by Prof. Yuliy Baryshnikov (University of Illinois at Urbana-Champaign)  Hyperbolic Geometry of Google Maps  at 2024-09-18 11:30:00
DESCRIPTION:Hyperbolic Geometry of Google Maps:Navigation of the Google Maps (not to confuse with *driving with* Google Maps) on smartphones is perhaps the most intuitive and efficient UI in existence, &ndash; and the reason, as I will show, is the underlying structure of the 3D hyperbolic space.Professor Yuliy Baryshnikov&rsquo;s short bio is here
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:861, Meyer Building
UID:eventx6a5a287eec3f010682
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20240918T143000
DTEND;TZID=Asia/Jerusalem:20240918T153000
DTSTAMP;TZID=Asia/Jerusalem:20240918T143000
SUMMARY: MSC  talk by Gilad Shmerler  Combinatorial Contracts with Constraints  at 2024-09-18 14:30:00
DESCRIPTION:The algorithmic study of the principal-agent framework is an emerging frontier for algorithmic game theory. We extend this model by incorporating knapsack constraints to capture real-world resource limitations. To address the computational challenges arising from these constraints, we develop approximation algorithms that guarantee near-optimal outcomes for both the principal and agents. Our research contributes to the understanding of contract design in complex environments.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:https://technion.zoom.us/j/2072669114?omn=95232443140 &amp; Taub 9
UID:eventx6a5a287eec3fa10673
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20240923T170000
DTEND;TZID=Asia/Jerusalem:20240923T220000
DTSTAMP;TZID=Asia/Jerusalem:20240923T170000
SUMMARY: CSpecial Event  Graduation Ceremony   at 2024-09-23 17:00:00
DESCRIPTION:Graduation Ceremony\nMonday, September 23, 2024\nLev HaCampus\n17:00 Technion Ceremony\n19:00 Gathering and Refreshments\n20:00 Ceremony Begins\nParking is available on a first-come, first-served basis.\nThe number of guests is limited to 4 due to space constraints.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Lev HaCampus
UID:eventx6a5a287eec40410680
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20240924T190000
DTEND;TZID=Asia/Jerusalem:20240924T220000
DTSTAMP;TZID=Asia/Jerusalem:20240924T190000
SUMMARY: CSpecial Event  *** Important Update *** Postponement of the Ceremony Dates.  at 2024-09-24 19:00:00
DESCRIPTION:According to the instructions from the Home Front Command, we regret to inform that we must postpone the excellence ceremony scheduled for Tuesday, September 24, 2024. We will provide further details later. We all hope for better and quieter days.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub
UID:eventx6a5a287eec40f10685
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20241009T103000
DTEND;TZID=Asia/Jerusalem:20241009T113000
DTSTAMP;TZID=Asia/Jerusalem:20241009T103000
SUMMARY: MSC  talk by Sapir Tubul  From Configuration-Space Clearance to Feature-Space Margin: Sample Complexity in Learning-Based Collision Detection  at 2024-10-09 10:30:00
DESCRIPTION:Motion planning is a central challenge in robotics, with learning-based approaches gaining significant attention in recent years. This thesis focuses on a specific aspect of these approaches: using machine-learning techniques, particularly Support Vector Machines (SVM), to evaluate whether robot configurations are collision-free, an operation termed "collision detection". Despite the growing popularity of these methods, there is a lack of theoretical guarantees supporting their efficiency and prediction accuracy. This is in stark contrast to the rich theoretical results of machine-learning methods in general and of SVMs in particular.\nThis work bridges this gap by analyzing the sample complexity of an SVM classifier for learning-based collision detection in motion planning. We provide a comprehensive theoretical framework that bounds the number of samples needed to achieve a specified accuracy at a given confidence level. This result is stated in terms relevant to robot motion-planning such as the system's clearance.\nBuilding on these theoretical results, we propose a collision-detection algorithm that can provide statistical guarantees on the algorithm's error in classifying robot configurations as collision-free or not. We rigorously prove the correctness of our approach and demonstrate its practical implications through extensive experimental evaluations.\nOur findings contribute to the theoretical understanding of learning-based collision detection and provide a foundation for developing more reliable and efficient motion planning algorithms. This work opens up new avenues for integrating machine learning techniques into robotics while maintaining provable performance guarantees.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:https://technion.zoom.us/j/2507428366
UID:eventx6a5a287eec41810683
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20241013T103000
DTEND;TZID=Asia/Jerusalem:20241013T113000
DTSTAMP;TZID=Asia/Jerusalem:20241013T103000
SUMMARY: PHD  talk by Idan Mehalel  Information and Randomness in Online Learning  at 2024-10-13 10:30:00
DESCRIPTION:Suppose that n forecasting experts are providing daily rain/no-rain predictions, and the best among them is mistaken in at most k many days. For how many days will an optimal learner allowed to observe the predictions mis-predict? This is a fundamental problem in online learning, and other important classification problems can be reduced to it. It was studied in the 90&rsquo;s by Cesa-Bianchi, Freund, Helmbold, and Warmuth who gave fine-grained bounds for the case where the learner may not use randomness. We study this problem without this restriction, and provide fine-grained bounds.There are many generalizations of this problem where the learner only receives partial feedback on its predictions. One suggested model is &ldquo;apple tasting feedback&rdquo;, in which the learner only knows the true outcome when it predicts a rainy day. Another very common model is &ldquo;bandit feedback&rdquo;, in which there are L &gt; 2 possible outcomes, and the learner only receives &ldquo;correct&rdquo; or &ldquo;incorrect&rdquo; feedback on its predictions. We study these models in various settings and provide nearly-tight mistake bounds.In this talk, we will discuss some of the techniques used to prove those mistake bounds. The mistake bound for the most classical full-information problem is proved via a reduction of the problem to calculating the (average) depth of binary trees, by using novel complexity measures of the set of experts. The bounds for the bandit feedback problem are proved by using minimax duality to obtain a dual variation of the problem, which is easier to handle.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 9
UID:eventx6a5a287eec42510679
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20241015T110000
DTEND;TZID=Asia/Jerusalem:20241015T120000
DTSTAMP;TZID=Asia/Jerusalem:20241015T110000
SUMMARY: MSC  talk by Yuval Lev Lubarsky  Generation and Application of Relational Database Embeddings  at 2024-10-15 11:00:00
DESCRIPTION:Machinery for data analysis often requires a numeric representation of the input. Towards that, a common practice is to embed components of structured data into a high-dimensional vector space. We study the embedding of the tuples of a relational database, where existing techniques are often based on optimization tasks over a collection of random walks from the database. The focus of this paper is on the recent FoRWaRD algorithm that is designed for dynamic databases, where walks are sampled by following foreign keys between tuples. Importantly, different walks have different schemas, or &ldquo;walk schemes,&rdquo; that are derived by listing the relations and attributes along the walk. Also importantly, different walk schemes describe relationships of different natures in the database.&nbsp;\nWe show that by focusing on a few informative walk schemes, we can obtain tuple embedding significantly faster, while retaining the quality. We define the problem of scheme selection for tuple embedding, devise several approaches and strategies for scheme selection, and conduct a thorough empirical study of the performance over a collection of downstream tasks. Our results confirm that with effective strategies for scheme selection, we can obtain high-quality embeddings considerably (e.g., three times) faster, preserve the extensibility to newly inserted tuples, and even achieve an increase in the precision of some tasks.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:https://technion.zoom.us/j/98306367803&nbsp;
UID:eventx6a5a287eec43110686
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20241031T110000
DTEND;TZID=Asia/Jerusalem:20241031T120000
DTSTAMP;TZID=Asia/Jerusalem:20241031T110000
SUMMARY: MSC  talk by Tal Haklay  A Framework for Automated Position-Aware Circuit Discovery  at 2024-10-31 11:00:00
DESCRIPTION:Mechanistic interpretability research seeks to explain the internal mechanisms that operate within AI models as they perform different tasks. A widely used strategy to uncover and analyze those mechanisms is by identifying circuits within the model. A circuit is a sub-graph of the model&rsquo;s computational graph, believed to be critical for executing a specific task.&nbsp;In recent years, several methods have been proposed for automatically identifying circuits inside languege models, but most overlook the timestep dimension. As a result, these approaches either produce circuits where nodes lack position specificity or are limited to datasets with uniform example lengths. In this work, we propose two key improvements: First, we extend edge attribution patching, a popular gradient-based method for circuit discovery, to differentiate between token positions. Second, we introduce the concept of a dataset schema&mdash;a structure that defines how to split examples in the dataset into spans based on shared structural features. By defining this high-level shared structure, we are able to discover circuits that differentiate between token positions, even in datasets that are not fully aligned. Additionally, we demonstrate that schema definition and span labeling can be automated using large language models (LLMs). Ultimately, our approach enables the fully automated discovery of position-sensitive circuits, yielding smaller circuits compared to previous methods.\n&nbsp;
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 8
UID:eventx6a5a287eec43d10689
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20241107T153000
DTEND;TZID=Asia/Jerusalem:20241107T163000
DTSTAMP;TZID=Asia/Jerusalem:20241107T153000
SUMMARY: MSC  talk by Meital Bojan  Correcting Flows with Marginal Matching  at 2024-11-07 15:30:00
DESCRIPTION:Flow matching models, ODE-based generative models, generate samples by gradually morphing a simple source distribution into a target distribution. In practice, these models still fall short of perfectly replicating the target distribution, mainly due to imperfections of the learned mapping. Previous work mainly focus on alleviating discretization error, which rises from sampling a continuous trajectory with a finite number of steps. In this work we focus on prediction error, an error that is inherent in the model. Our main contribution is identifying a trajectory that complies with the imperfect flow model and leads exactly to the target distribution. Based on this finding, we propose Marginal Matching---a simple inference-time correction scheme to steer the generated samples in the direction of the data. This scheme proves to reduce a bound on the distance between the data and the learned distribution, motivating two different implementations for the correction function. We show that our proposed method improves sample quality on CIFAR-10 and ImageNet-64, &nbsp;with minimal overhead in computation time, or non at all when applying approximated correction.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:https://technion.zoom.us/j/3847521859
UID:eventx6a5a287eec44910687
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20241112T113000
DTEND;TZID=Asia/Jerusalem:20241112T123000
DTSTAMP;TZID=Asia/Jerusalem:20241112T113000
SUMMARY: pixel-club  talk by Niv Nayman   Advancing Automatic Machine Learning (AutoML)  at 2024-11-12 11:30:00
DESCRIPTION:Machine learning (ML) has achieved considerable successes in recent years and an ever-growing number of disciplines rely on it. However, this success crucially relies on human experts manually processing data to build, select and train models. In this work we focus on advancing two different aspects of Automated Machine Learning (AutoML) and provide automatic methods and processes to mitigate some of the reliance on ML experts.First, we introduce an interpretable neural architecture search method to efficiently optimize a deep neural network under latency constraints on different devices:Making neural networks practical often requires adhering to resource constraints such as latency, energy and memory. We solve this by introducing a Bilinear Interpretable approach for constrained Neural Architecture Search (BINAS). Our method is based on an accurate and simple bilinear formulation of both an accuracy estimator and the expected resource requirement, jointly with a scalable search method with theoretical guarantees. One major advantage of BINAS is providing interpretability via insights about the contribution of different design choices. For example, we find that in the examined search space, adding depth and width is more effective at deeper stages of the network and at the beginning of each resolution stage. BINAS differs from previous methods that typically use complicated accuracy predictors that make them hard to interpret, sensitive to many hyper-parameters, and thus with compromised final accuracy. Our experiments show that BINAS generates comparable to or better than state of the art architectures, while reducing the marginal search cost, as well as strictly satisfying the resource constraints. Secondly, we present an extensive study for identifying intrinsic properties of pre-trained models for model selection in the context of transfer learning to different downstream tasks. We quesiton the commonly accepted hypothesis stating that models with higher accuracy on Imagenet perform better on other downstream tasks, which led to much research dedicated to optimizing Imagenet accuracy. Recently this hypothesis has been challenged by evidence showing that self-supervised models transfer better than their supervised counterparts, despite their inferior Imagenet accuracy. This calls for identifying the additional factors, on top of Imagenet accuracy, that make models transferable. In this work we show that high diversity of the filters learnt by the model promotes transferability jointly with Imagenet accuracy. Encouraged by the recent transferability results of self-supervised models, we use a simple procedure to combine self-supervised and supervised pretraining and generate models with both high diversity and high accuracy, and as a result high transferability. We experiment with several architectures and multiple downstream tasks, including both single-label and multi-label classification.\nNiv Nayman is an applied scientist at Amazon Web Services AI Labs and a Phd candidate at the Technion, working primarily on AutoML research with applications in computer vision and document analysis. His work has been integrated in products at scale and published at top venues (NeurIPS, ICLR, ICML, ECCV, etc.). Before joining Amazon, Niv served as a senior research scientist at Alibaba DAMO Academy after completing a long service as an officer in the intelligence technological unit &ndash; 81, where he performed a variety of roles through the years, from hardware design, to algorithmic research and cyber security. Niv holds BSc degrees both in Electrical Engineering and in Physics (Cum Laude, &lsquo;Psagot&rsquo; program) and a MSc in Optimization and Machine Learning, all from the Technion.\nPh.D. &nbsp;student under the supervision of Prof. Lihi Zelnik Manor.\n(Andrew and Erna Viterbi Faculty of Electrical &amp; Computer Engineering)
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:https://technion.zoom.us/j/91527524137
UID:eventx6a5a287eec45510690
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20241117T183000
DTEND;TZID=Asia/Jerusalem:20241117T200000
DTSTAMP;TZID=Asia/Jerusalem:20241117T183000
SUMMARY: CSpecial Event  Intel AI Workshop with Dana Israeli  at 2024-11-17 18:30:00
DESCRIPTION:You are Invited to Intel's practical Gen AI workshop.\nRegistration:\nלינק לסדנא של ה- 17\nלינק לסדנא של ה- 20\n&nbsp;
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Virtual
UID:eventx6a5a287eec46710688
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20241120T110000
DTEND;TZID=Asia/Jerusalem:20241120T120000
DTSTAMP;TZID=Asia/Jerusalem:20241120T110000
SUMMARY: MSC  talk by Amit Zrihan  Studying the Cycle Complexity of DNA Synthesis  at 2024-11-20 11:00:00
DESCRIPTION:DNA data storage presents an efficient solution for archiving, though synthesis time and cost pose challenges.This seminar focuses on cyclic synchronized synthesis technologies like photolithography, introducing performance metrics based on synthesis cycles.We extend prior work on channel capacity, achieving higher rates and capacities through improved encoding.Additionally, we analyze cost bounds and explore alphabet sizes larger than the standard four, inspired by recent advancements in DNA synthesis methods.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 601
UID:eventx6a5a287eec47510684
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20241120T124500
DTEND;TZID=Asia/Jerusalem:20241120T134500
DTSTAMP;TZID=Asia/Jerusalem:20241120T124500
SUMMARY: Theory Semina  talk by Omri Ben-Eliezer (Technion)  Approximate counting of permutation patterns  at 2024-11-20 12:45:00
DESCRIPTION:A copy of a permutation pattern (say, 132) in a sequence of numbers is any subsequence whose values have the same relative order as in the pattern. (Say, for 132, the first element is smallest, the second is largest, and the third is in-between.).&nbsp;Counting permutation patterns has a surprisingly rich set of connections and applications in ranking, statistics, combinatorics, fine-grained complexity, and parametrized complexity, especially for fixed small k. Here are three examples:(i) Counting 4-cycles in sparse graphs is equivalent to counting 4-patterns [Dudek and Gawrychowski, 2020].(ii) Many fundamental tests in nonparametric statistics amount to counting k-patterns for k up to 5.(iii) The study of twin-width in parametrized complexity has originated from a breakthrough FPT algorithm of Guillemot and Marx [2013] for permutation pattern detection, which runs in linear time when k is fixed.&nbsp;In this talk I will describe an algorithm for approximately counting all k-patterns for k up to 5 in near-linear time, deterministically, to within a (1+eps)-multiplicative error. This algorithm gives the first known (conditional) separation between exact and approximate counting in this domain. Interestingly, our algorithm leverages sublinear techniques from distribution testing, which to our knowledge have not been used in a pattern counting context before.&nbsp;Joint work with Slobodan Mitrović (UC Davis) and Pranjal Srivastava (MIT).
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 401
UID:eventx6a5a287eec47f10692
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20241126T183000
DTEND;TZID=Asia/Jerusalem:20241126T203000
DTSTAMP;TZID=Asia/Jerusalem:20241126T183000
SUMMARY: CSpecial Event  Come be Part of the Faculty's CTF Group - Meeting on November 26th  at 2024-11-26 18:30:00
DESCRIPTION:Come be part of the Technion's Capture The Flag-CTF group!!&nbsp;\nThe meeting will take place on Tuesday 26.11 at 18:30 in Taub 8 for beginners and 201 for advanced.\nThis week we will learn about weaknesses in the Web! How can you make a website show you information that you are not supposed to see?\nEveryone is invited, even if you did not come to the opening meeting! Looking forward to seeing you\nTo register: https://forms.gle/ZPC4U4arB4pgu5vVA&nbsp;\nCommunity link: https://chat.whatsapp.com/CfBxBd9yoc070pgFNF0S4F
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 8 / Room 201
UID:eventx6a5a287eec48b10693
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20241128T143000
DTEND;TZID=Asia/Jerusalem:20241128T163000
DTSTAMP;TZID=Asia/Jerusalem:20241128T143000
SUMMARY: colloq  talk by David Tennenhouse, Senior Advisor, National Science Foundation  AI Infrastructure: We CAN Close the AI Energy Gap  at 2024-11-28 14:30:00
DESCRIPTION:According to some pundits, growth in the demand for AI training and inference so vastly exceeds Moore&rsquo;s Law that the demand can only be met with a rapidly growing population of ever more power-hungry data centers that could collectively consume a significant fraction of the world&rsquo;s electricity. Our job, as the visionaries and engineers of the Infrastructure enabling AI, is to ensure that disproportionate surge in energy demand doesn&rsquo;t happen, i.e., we must deliver the benefits of AI in ways that are sustainable and economically viable. This presentation &nbsp;will begin with a realistic discussion around the &ldquo;demand&rdquo; side of the equation. It will then identify an offsetting set of opportunities for innovation on the &ldquo;supply&rdquo; side &mdash; opportunities to so vastly improve the effectiveness of AI infrastructure that, in aggregate, they can offset the growth on the demand side. The supply side discussion will highlight examples across the infrastructure &ldquo;stack&rdquo;: in the models, the algorithms, the software, the individual compute nodes and in the scale-out mechanisms used at the data center level.\n&nbsp;Bio:\nDavid is passionate about research and innovation and has a track record of embracing high-risk initiatives, such as software-defined networking, software radio, IoT, and data-intensive computing. He has worked in academia, as a faculty member at MIT; in government, at DARPA and NSF; in industry at Intel, Amazon/A9.com, Microsoft, and VMware; and as a partner in a venture capital firm. Dr. Tennenhouse has championed research related to a wide range of technologies, including networking, distributed computing, blockchain, computer architecture, storage, machine learning, robotics, and nano/biotechnology. David holds a BASc and MASc in Electrical Engineering from the University of Toronto and obtained his Ph.D. at the University of Cambridge.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub Piano Auditorium (012)
UID:eventx6a5a287eec49710695
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20241203T183000
DTEND;TZID=Asia/Jerusalem:20241203T203000
DTSTAMP;TZID=Asia/Jerusalem:20241203T183000
SUMMARY: CSpecial Event  Come be Part of the Faculty's CTF Group - Meeting on December 3rd  at 2024-12-03 18:30:00
DESCRIPTION:Come be part of the Technion's Capture The Flag-CTF group!!\nThe third meeting will be held on Tuesday, December 3rd at 6:30 PM in Taub 6 for beginners and 8 for advanced.\nThis week in Beginners we will continue to learn about vulnerabilities in the Web! And this time XSS. How can you make a website run any code you want?In Advanced we will deal with Server side vulnerabilities, present examples and solve Web challenges\nEveryone is welcome, even if you haven't come to previous meetings!Looking forward to seeing you\nRegister at the link\nCommunity link
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 8 / Taub 6
UID:eventx6a5a287eec4a510700
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20241204T123000
DTEND;TZID=Asia/Jerusalem:20241204T143000
DTSTAMP;TZID=Asia/Jerusalem:20241204T123000
SUMMARY: CSpecial Event  Spotlight Day with Apple – Don't miss it! 4.12 at the IDF – Faculty of Electrical Engineering  at 2024-12-04 12:30:00
DESCRIPTION:This Wednesday, Apple's recruiting team and engineers will be at Spotlight Day at the Faculty of Electrical Engineering.This is a unique opportunity to hear about groundbreaking technologies and meet the people behind the scenes!\nOn the program:- 12:30 Lecture by Assaf Menachem\nHow custom silicon made Apple Vision Pro possible\nLocation: Room 354, 3rd floor, Meyer Building\n- 13:00 Meeting with the recruiting team and engineers from the company\nLocation: 3rd floor lobby, Meyer Building\nAnd no less important:So that you don't stay hungry - surprises and Hot Potatoes await you!\nMark your calendars and come!\nRegistration link here\nWe are waiting for you!
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Room 354, 3rd floor, Meyer Building
UID:eventx6a5a287eec4b010701
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20241204T130000
DTEND;TZID=Asia/Jerusalem:20241204T140000
DTSTAMP;TZID=Asia/Jerusalem:20241204T130000
SUMMARY: Theory Semina  talk by Shiri Ron (Weizmann)  Theory Seminar: Revisiting Combinatorial Auctions: Navigating a Hierarchy of "Truthfulness" Notions  at 2024-12-04 13:00:00
DESCRIPTION:In the basic setting of algorithmic mechanism design, the goal is to design an algorithm that solves an optimization problem whose input is distributed among rational agents. We assume that the output of the algorithm matters to the agents, so game theoretic considerations should be taken into account. An example that effectively illustrates this scenario is provided by auctions. We focus on auctions where the goal is to maximize the collective value for all participating parties, an objective formally known as the social welfare.\nIn this talk, we ask: Can we design auctions that satisfy three key desiderata &ndash; social welfare optimization, robustness to strategic manipulation (also termed &ldquo;truthfulness&rdquo; and incentive compatibility) and polynomial communication complexity? Despite ample effort, the answer to this question has remained elusive. We note that most of the attention has been directed towards one means of robustness against strategic manipulation, namely towards mechanisms that can be implemented in a Nash equilibrium, namely ex-post incentive compatible mechanisms.\nWe consider two additional properties of mechanisms that capture robustness against strategic manipulation. We start by considering the more restricted class of dominant-strategy mechanisms, which are mechanisms that can be implemented in a dominant-strategy equilibrium. We show several impossibilities for them.\nWe then examine the class of obviously strategy-proof mechanisms. This class was introduced by Li [American Economic Review &rsquo;17] and it describes mechanisms with dominant strategies that are also self-explanatory. We show that even for bidders with simple valuations such as additive and unit-demand valuations, obviously strategy-proof mechanisms are powerless.\nThe talk is based on joint works with Shahar Dobzinski, Dan Schoepflin and Jan Vondr&aacute;k. 
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 401
UID:eventx6a5a287eec4bc10698
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20241204T170000
DTEND;TZID=Asia/Jerusalem:20241204T180000
DTSTAMP;TZID=Asia/Jerusalem:20241204T170000
SUMMARY: MSC  talk by Niv Kaminer  Static Huge-Page Allocation Guided by Execution Profiles  at 2024-12-04 17:00:00
DESCRIPTION:Dynamic algorithms to construct huge pages can improve the performance of big-memory workloads but might introduce latencies during page construction. Some workloads may benefit from static algorithms that utilize offline profiling information to predetermine memory allocations. But while such algorithms are more predictable, they are: (1) unsuitable for setups with limited physical memory contiguity; (2) ineffective whenever program call sites allocating memory are invoked via different code paths; and (3) vulnerable to minor configuration and input changes.\nWe propose Salloc, a profile-guided static allocator that addresses these drawbacks by: (1) prioritizing huge page allocations that matter most; (2) defining allocation sites using their full call stack; and (3) cumulatively supporting multiple offline execution profiles. We show Salloc is effective across similar application datasets and different contiguity limitations. For example, it achieves up to 90% of the maximal speedup when backing only 5% of application memory with huge pages (rather than all application memory). Nevertheless, we find that static huge page allocation is inevitably ineffective if different program inputs trigger substantially different allocation policies within the application, because the offline profile might align with a different policy than the running program.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 601
UID:eventx6a5a287eec4cb10696
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20241205T103000
DTEND;TZID=Asia/Jerusalem:20241205T113000
DTSTAMP;TZID=Asia/Jerusalem:20241205T103000
SUMMARY: MSC  talk by Diana Cohen  Distributed Recoverable Sketches  at 2024-12-05 10:30:00
DESCRIPTION:Sketches are commonly used in computer systems and network monitoring tools to provide efficient query executions while maintaining a compact data representation.&nbsp;Switches and routers maintain sketches to track statistical characteristics of the network traffic.&nbsp;The availability of such data is essential for the network analysis as a whole.&nbsp;Consequently, being able to recover sketches is critical following a switch crash.&nbsp;In this work, we explore how nodes in a network environment can cooperate to recover sketch data whenever any subset of them crashes.&nbsp;We consider various approaches to ensure data reliability and explore the trade-offs between space consumption, runtime overheads, and traffic during recovery.&nbsp;A key aspect we examine is how nodes update each other about their sketch content as it evolves over time.&nbsp;In particular, we focus on periodic incremental updates using buffering and batching techniques.&nbsp;We also examine several data structures to economically represent and encode a batch of latest changes.&nbsp;
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 8
UID:eventx6a5a287eec4d610697
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20241205T140000
DTEND;TZID=Asia/Jerusalem:20241205T150000
DTSTAMP;TZID=Asia/Jerusalem:20241205T140000
SUMMARY: PHD  talk by Asaf Yeshurun  Simplification of Counter Automata  at 2024-12-05 14:00:00
DESCRIPTION:Quantitative computational models have become increasingly popular in the past few decades, since they can naturally model distributed systems and enable reasoning about quantitative properties and about infinite configuration spaces. Specific models include Petri nets, Vector Addition Systems (VAS\VASS), Weighted Automata and Counter Nets. Their usefulness lies in the fact that they retain decidability of some important decision problems.Nonetheless, the complexities of reasoning about these models are typically very high. Thus, it is desirable to simplify these models as much as possible.&nbsp;Simplification can take many forms, e.g., determinization, dimension reduction, &nbsp;and many more.In this lecture I give an overview of these models, their decision problems, and our works regarding various types of simplification on them.Specifically, I will discuss the determinization process of OCNs (one-counter nets), detail how various notions of determinization arise naturally from the intricacies of the model, and prove complexity results for the determinization decision problems that arise. I also introduce and explore the notion of history-determinism - a restriction of determinism with its own merit and relevance.I will also introduce the notions of dimension-minimality and primality of counter nets, notions tightly linked to the concepts of minimization and simplification, and show that primality is undecidable.Finally, I will discuss two-way OCNs (2-OCNs) -- a stronger variant of OCNs. I introduce new results of its expressive power under various restrictions, and shed light on interesting links between 2-OCNs and semilinear languages.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:https://technion.zoom.us/j/6919019730&nbsp; &amp; Taub 601
UID:eventx6a5a287eec4e110691
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20241211T130000
DTEND;TZID=Asia/Jerusalem:20241211T140000
DTSTAMP;TZID=Asia/Jerusalem:20241211T130000
SUMMARY: Theory Semina  talk by Talya Eden (Bar-Ilan University)  Theory Seminar: Approximately counting and sampling Hamiltonian motifs in sublinear-time  at 2024-12-11 13:00:00
DESCRIPTION:Counting small subgraphs, known as motifs, within large graphs is a fundamental challenge in graph analysis. In this talk, I will introduce a novel algorithm in the standard query model for approximately counting and almost uniformly sampling any Hamiltonian motif in sublinear time. Our approach highly simplifies prior approximate counting algorithms for edges, cliques, and stars in the standard model. This work marks a significant step toward narrowing the gap between results achievable in the standard model and the more powerful augmented query model.\nJoint work with Reut Levi, Dana Ron and Ronitt Rubinfeld
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 401
UID:eventx6a5a287eec4ee10703
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20241212T110000
DTEND;TZID=Asia/Jerusalem:20241212T130000
DTSTAMP;TZID=Asia/Jerusalem:20241212T110000
SUMMARY: pixel-club  talk by Yiftach Edelstein    Pixel Club - Sharp-It: A Multi-view to Multi-view Diffusion Model for 3D Synthesis and Manipulation  at 2024-12-12 11:00:00
DESCRIPTION:Advancements in text-to-image diffusion models have led to significant progress in fast 3D content creation. One common approach is to generate a set of multi-view images of an object, and then reconstruct it into a 3D model. However, this approach bypasses the use of a native 3D representation of the object and is hence prone to geometric artifacts and limited in controllability and manipulation capabilities. An alternative approach involves native 3D generative models that directly produce 3D representations. These models, however, are typically limited in their resolution, resulting in lower quality 3D objects. In this work, we bridge the quality gap between methods that directly generate 3D representations and ones that reconstruct 3D objects from multi-view images. We introduce a multi-view to multi-view diffusion model called Sharp-It, which takes a 3D consistent set of multi-view images rendered from a low-quality object and enriches its geometric details and texture. The diffusion model operates on the multi-view set in parallel, in the sense that it shares features across the generated views. A high-quality 3D model can then be reconstructed from the enriched multi-view set. By leveraging the advantages of both 2D and 3D approaches, our method offers an efficient and controllable method for high-quality 3D content creation. We demonstrate that Sharp-It enables various 3D applications, such as fast synthesis, editing, and controlled generation, while attaining high-quality assets.\nM.Sc. student under the supervision of Prof. Lihi Zelnik &ndash; Manor.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Room 1061, Meyer Building
UID:eventx6a5a287eec4f910705
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20241216T173000
DTEND;TZID=Asia/Jerusalem:20241216T203000
DTSTAMP;TZID=Asia/Jerusalem:20241216T173000
SUMMARY: CSpecial Event  SHE-S Technical Job Interview Preparation Workshop  at 2024-12-16 17:30:00
DESCRIPTION:SHE-S, the student community of Computer Science, is pleased to invite you to a technical job interview preparation workshop that will give you practical tools to stand out in the recruitment process and pass the interviews.\nWhen? Monday, December 16, at 5:30 PM\nWhere? Auditorium Piano in Taub (floor 0)\nWhat will we talk about?\n&middot;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;What a technical interview looks like and how to prepare for it effectively\n&middot;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;Tools for dealing with challenging questions: how to answer, and what to do when you get stuck\n&middot;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;Practice real questions from job interviews in the industry\n&middot;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp;Presenting projects in an impressive and convincing way\n&middot;&nbsp;&nbsp;&nbsp;&nbsp;&nbsp; And other unique tips that will help you stand out from the candidates\nWho will be the moderator?\nAdi Mano &ndash; a manager and technical lead at Google with extensive experience as an interviewer and someone who accompanies candidates in the recruitment process.\nWhy should you participate?In a challenging job market, where there are fewer opportunities and more candidates, good preparation can make all the difference.\nThe workshop is designed to help you arrive at interviews prepared, with self-confidence and the ability to impress the interviewers.\nWho is the workshop suitable for?For students looking for student jobs and also for those at the beginning of their technological career (junior).\nTo register: Here &ndash; the number of places is limited!\nDon't wait &ndash; this is your opportunity to prepare, stand out and lead the job application process in a challenging and dynamic market.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Piano Auditorium, Floor 0 (012)
UID:eventx6a5a287eec50510699
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20241217T183000
DTEND;TZID=Asia/Jerusalem:20241217T203000
DTSTAMP;TZID=Asia/Jerusalem:20241217T183000
SUMMARY: CSpecial Event  Come be part of the faculty's CTF group - meeting on December 17th  at 2024-12-17 18:30:00
DESCRIPTION:Come be part of the Technion's Capture The Flag-CTF group!!\nAnd this week, a guest lecture on crypto, on secure computing in the cloud and how it relates to cryptography and artificial intelligence\nThe meeting will take place on Tuesday, December 17 at 6:30 PM in Taub 3.\nAfter the lecture, we will solve CTF challenges on cryptography\nEveryone is welcome, even if you haven't been to previous meetings!\nYou must register in advance at the following link:https://forms.gle/bVDdVCyFRMNWty876\nWe look forward to seeing you
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 3
UID:eventx6a5a287eec51810708
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20241218T123000
DTEND;TZID=Asia/Jerusalem:20241218T143000
DTSTAMP;TZID=Asia/Jerusalem:20241218T123000
SUMMARY: CSpecial Event  Spotlight Day with NVIDIA  at 2024-12-18 12:30:00
DESCRIPTION:You are invited to a Spotlight Day with NVIDIAWednesday, December 18, starting at 12:30At the Faculty of Computer Science, Taub Building\nWhat's on the program?Mingling with the recruitment teams and company engineers in the Taub LobbyA fascinating lecture and professional panel on the topic: "5G RAN and AI." Speaker - Elran Avisror, Senior Director of the Software Group at NVIDIA Israel. At Taub 1\nRefreshments are in order!\nThis is an excellent opportunity to get to know the people behind the world's advanced technologies, hear about technological innovation, and discover fascinating career opportunities at NVIDIA\nRegister in advance at the link here\nEveryone is invited!
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 1
UID:eventx6a5a287eec52210706
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20241218T130000
DTEND;TZID=Asia/Jerusalem:20241218T140000
DTSTAMP;TZID=Asia/Jerusalem:20241218T130000
SUMMARY: Theory Semina  talk by Shay Solomon (Tel-Aviv University)  Theory Seminar: Vizing's Theorem in Near-Linear Time  at 2024-12-18 13:00:00
DESCRIPTION:Vizing&rsquo;s Theorem from 1964 states that any n-vertex m-edge graph of maximum degree &Delta; can be edge colored using at most &Delta;+1 different colors. Vizing&rsquo;s original proof is algorithmic and implies that such an edge coloring can be found in O(mn) time. In this talk, I&rsquo;ll present a randomized algorithm that computes a (&Delta;+1)-edge coloring in near-linear time &mdash; in fact, only O(mlog&Delta;) time &mdash; with high probability.\nBased on&nbsp;https://arxiv.org/abs/2410.05240
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 4
UID:eventx6a5a287eec54110707
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20241230T100000
DTEND;TZID=Asia/Jerusalem:20241230T120000
DTSTAMP;TZID=Asia/Jerusalem:20241230T100000
SUMMARY: CSpecial Event  Interactive Race for the Future Generation - Following Encryption and Cyber ​​in Collaboration with the Technion Alumni Organization and the Faculty of Biomedical Engineering  at 2024-12-30 10:00:00
DESCRIPTION:It's time to go out, brothers-uncles-parents who are hardworking!!\nThis coming Hanukkah, we are opening the faculty for an interactive race following encryption and cyber together with the Technion Alumni Organization and the Faculty of Biomedical Engineering.\nOn the program: a race following ciphers and encryption, an introduction to MRI systems and a visit to an innovative laboratory.\nMonday, December 30th starting at 10:00\nFor details and registration, visit the link (limited number of places)\nThe activity is suitable for ages 16 and up.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Technion
UID:eventx6a5a287eec55410710
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20250101T113000
DTEND;TZID=Asia/Jerusalem:20250101T133000
DTSTAMP;TZID=Asia/Jerusalem:20250101T113000
SUMMARY: colloq  talk by Tomer Ezra  Algorithmic Contract Design  at 2025-01-01 11:30:00
DESCRIPTION:We explore the framework of contract design through a computational perspective. Contract design is a fundamental pillar of microeconomics, addressing the essential question of how to incentivize people to work. The significance of contract design was acknowledged by the Nobel Prize awarded to Hart and Holmstr&ouml;m, and it applies to various real-life scenarios, such as determining bonuses for employees, setting commission structures for sales representatives, and designing payment schemes for influencers promoting products.\nWhile contract design has been extensively studied from an economic perspective, this talk will examine it from a computational viewpoint. Specifically, we introduce combinatorial extensions of classic contract design models, where a principal delegates tasks to one or multiple agents. The agents have sets of potential actions they can take to complete the task, and the chosen actions by the agents stochastically determine the success of the task. We analyze the structure and computational aspects of these models, and present algorithms that provide (approximately) optimal guarantees.\nShort Bio:Tomer Ezra is a postdoctoral fellow at the Center of Mathematical Sciences and Applications (CMSA) at Harvard University. Previously, he was a Sloan Postdoctoral Fellow at the Simons Laufer Mathematical Sciences Institute (SLMath) and a Postdoctoral Fellow at Sapienza University of Rome, hosted by Prof. Stefano Leonardi. He earned his PhD from Tel Aviv University, advised by Prof. Michal Feldman. His research lies at the intersection of computer science and economics, with a focus on analyzing and designing simple mechanisms and algorithms in limited information environments.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Room 337
UID:eventx6a5a287eec56410712
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20250101T163000
DTEND;TZID=Asia/Jerusalem:20250101T183000
DTSTAMP;TZID=Asia/Jerusalem:20250101T163000
SUMMARY: colloq  talk by Dr. Tamar Eilam (IBM)  The AI Energy Problem and What Can We Do About It  at 2025-01-01 16:30:00
DESCRIPTION:Artificial intelligence (AI) offers immense potential to accelerate scientific discoveries crucial for combating climate change. However, this powerful tool comes with a significant environmental cost due to its substantial energy consumption and carbon emissions. This talk explores the research challenge of harnessing AI's capabilities while minimizing its ecological footprint.&nbsp;Bio: Dr. Tamar Eilam is an IBM Fellow and Chief Scientist for Sustainable Computing in the IBM T. J. Watson Research Center, New York. &nbsp;Tamar is leading research aiming at drastically reducing the carbon footprint associated with computing across infrastructure, systems, and software, data and AI. Tamar completed a Ph.D. degree in Computer Science in the Technion, Israel, in 2000. She joined the IBM T.J. Watson Research Center in New York as a Research Staff Member that same year. She was recognized as an IBM Fellow in 2014.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 9
UID:eventx6a5a287eec57510714
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20250106T103000
DTEND;TZID=Asia/Jerusalem:20250106T123000
DTSTAMP;TZID=Asia/Jerusalem:20250106T103000
SUMMARY: colloq  talk by Yanai Elazar (The University of Washington)  Understanding Generative Models Inside Out: From Representation to Data  at 2025-01-06 10:30:00
DESCRIPTION:Generative models, such as ChatGPT and DALL-E, are used by millions of people daily for tasks ranging from programming and content creation to resume filtering. These models often create the impression of being &ldquo;intelligent,&rdquo; which can incentivize careless use in critical applications. While generative models are empowering, they appear to be black boxes, and their misuse can result in harmful or unlawful outcomes.\nIn this talk, I will present algorithms and tools for dissecting and analyzing generative models using holistic, causal, and data-centric approaches.By applying these methods to state-of-the-art models, we can foster trust in these technologies by uncovering human-interpretable concepts that underpin their behavior, scrutinizing their extensive training data, and evaluating their learning processes.\nFinally, I will reflect on how generative models have transformed the field of AI and discuss the challenges that remain in ensuring their responsible development and use.\nBio: Yanai Elazar is a Postdoctoral Researcher at AI2 and the University of Washington. Prior to that, he completed his PhD in Computer Science at Bar-Ilan University. He is interested in the science of generative models, for which he develops algorithms and tools for understanding what makes models work, how, and why.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Room 337
UID:eventx6a5a287eec58510715
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20250106T130000
DTEND;TZID=Asia/Jerusalem:20250106T140000
DTSTAMP;TZID=Asia/Jerusalem:20250106T130000
SUMMARY: MSC  talk by Gal Yona  Self-Supervised Learning of Robust Local Surface Descriptors Using Polynomial Patches  at 2025-01-06 13:00:00
DESCRIPTION:Classical shape descriptors such as Heat Kernel Signature (HKS), Wave Kernel Signature (WKS), and Signature of Histograms of Orientations (SHOT), while widely used in shape analysis, exhibit sensitivity to mesh connectivity, sampling patterns, and topological noise.\nWhile differential geometry offers a promising alternative through its theory of differential invariants, which are theoretically guaranteed to be robust shape descriptors, the computation of these invariants on discrete meshes often leads to unstable numerical approximations, limiting their practical utility. We present a self-supervised learning approach for extracting geometric features from 3D surfaces. Our method combines synthetic data generation with a neural architecture designed to learn sampling-invariant features.\nBy integrating our features into existing shape correspondence frameworks, we demonstrate improved performance on standard benchmarks including FAUST, SCAPE, TOPKIDS, and SHREC'16, showing particular robustness to topological noise and partial shapes.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 401 &amp; Zoom:&nbsp;94964568766
UID:eventx6a5a287eec59610709
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20250107T103000
DTEND;TZID=Asia/Jerusalem:20250107T123000
DTSTAMP;TZID=Asia/Jerusalem:20250107T103000
SUMMARY: colloq  talk by Leshem Choshen (MIT-IBM)  Communal AI -  Open, Collaborative & Accessible LLMs  at 2025-01-07 10:30:00
DESCRIPTION:Developing better Language Models would benefit a myriad of communities. However, it is prohibitively costly. The talk would describe collaborative approaches to pretraining such as model merging, allowing combining several specialized models into one. Then introduce efficient evaluation to reduce overheads and touch on other accessible and collaborative aspects that best harness the expertise and diversity in Academia.Bio: Leshem Choshen is a postdoctoral researcher at MIT&amp;IBM, aiming to study model development openly &amp; collaboratively, allow feasible pretraining research, and evaluate efficiently. To do so they co-created model merging, TIES merging, and the babyLM challenge. They were chosen for the postdoctoral Rothschild and Fulbright fellowship as well as IAAI and Blavatnik best Ph.D. awards. With broad NLP and ML interests, they also worked on Reinforcement Learning, Understanding how neural networks learn, and the Nature cover Project Debater &ndash; the first (2019) machine to hold a formal debate (live). Leshem is also a dancer and an acrobat.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Room 337
UID:eventx6a5a287eec5a610717
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20250107T183000
DTEND;TZID=Asia/Jerusalem:20250107T203000
DTSTAMP;TZID=Asia/Jerusalem:20250107T183000
SUMMARY: CSpecial Event  Come Be Part Of The Faculty's CTF Group - Meeting On January 7th  at 2025-01-07 18:30:00
DESCRIPTION:Come be part of the Technion's Capture The Flag-CTF group!\nAnd this week: a guest lecture on the topic of artificial intelligence security! Building and Breaking AI Security, hosted by Amit Levy and Rom Himmelstein &ndash; AI security researchers at the Technion\n&ldquo;The Model They Told You Not to Worry About&rdquo; How weaknesses in language models can cause systems to crash and even expose personal information?\nDon't have previous experience with language models? Don't worry! The lecture will be practical, and in 7 minutes you can arrive prepared with the help of the video at the link\nTuesday, 7.1, at 18:30Taub 3Register at the link\nAfter the lecture, we will solve CTF challenges on the topic together! Everyone is invited, even if you haven't come to previous meetings!See you there!________________________________________
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 3
UID:eventx6a5a287eec5b710720
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20250108T123000
DTEND;TZID=Asia/Jerusalem:20250108T141500
DTSTAMP;TZID=Asia/Jerusalem:20250108T123000
SUMMARY: CSpecial Event  TII Internship Spotlight Day  at 2025-01-08 12:30:00
DESCRIPTION:Join us for our Internship Spotlight Day, where we&rsquo;ll introduce the TII AI/IR Research Center recently established in Haifa, discuss our work in Generative AI and share details about our 2025 Internship Program.\nDate: Wednesday, January 8&nbsp;Time: 12:30&ndash;14:15&nbsp;Agenda:&nbsp;12:30-13:00: Get-together&nbsp;13:00-13:50 : Introduction &amp; Tech Talk &nbsp;&ldquo;The Power of Noise: unexpected findings in Retrieval Augmented Generation&rdquo; (SIGIR&rsquo;2024)13:50-14:15 Meet the TII researchers and discuss opportunities for collaboration &amp; summer research internships.&nbsp;We look forward to seeing you there!Please RSVP in this registration form by Jan 1stFor any inquiry, please send mail to &lt;michal.caspi@tii.ae&gt;&nbsp;* The event is by invitation only and restricted to enrolled graduate students in CS/EE/DDS and faculty members.Snacks will be served. Early registered attendees will be eligible for swag (on first come first served basis)
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Graduate launge, 2nd floor
UID:eventx6a5a287eec5c610711
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20250108T130000
DTEND;TZID=Asia/Jerusalem:20250108T140000
DTSTAMP;TZID=Asia/Jerusalem:20250108T130000
SUMMARY: Theory Semina  talk by Tomer Adar (Technion)  Theory Seminar: Support Testing in the Huge Object Model  at 2025-01-08 13:00:00
DESCRIPTION:The Huge Object model is a distribution testing model in which we are given access to independent samples from an unknown distribution over the set of strings {0,1}^n, but are only allowed to query a few bits from the samples. We investigate the problem of testing whether a distribution is supported on m elements in this model. It turns out that the behavior of this property is surprisingly intricate, especially when also considering the question of adaptivity. We prove lower and upper bounds for both adaptive and non-adaptive algorithms in the one-sided and two-sided error regime. Our bounds are tight when m is fixed to a constant (and the distance parameter &epsilon; is the only variable). For the general case, our bounds are at most O(log m) apart. In particular, our results show a surprising O(log 1/&epsilon;) gap between the number of queries required for non-adaptive testing as compared to adaptive testing. For one sided error testing, we also show that a O(log m) gap between the number of samples and the number of queries is necessary. Our results utilize a wide variety of combinatorial and probabilistic methods.\nJoint work with Eldar Fischer and Amit Levi (https://arxiv.org/pdf/2308.15988)
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 4
UID:eventx6a5a287eec5da10718
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20250108T183000
DTEND;TZID=Asia/Jerusalem:20250108T203000
DTSTAMP;TZID=Asia/Jerusalem:20250108T183000
SUMMARY: CSpecial Event  "Research on the Bar" Evening January 8, 2025  at 2025-01-08 18:30:00
DESCRIPTION:You are invited to the evening of "Research on the Bar" - three faculty members giving TED talks at eye level.\nDon't miss the opportunity to get to know the researchers and new research groups, in an open atmosphere with beers and snacks.\nWednesday, January 8, 2025 starting at 6:30 PM at Taub 2.\n\nDr. Brit Youngmann -&nbsp;Put your trust in the data (and not in the person who processed it)\nDr. Or Litany - Lights, Camera, Action! Revolutionizing Computer Vision with Video Generative AI\nDr. Gala Yadgar - The picture of Schr&ouml;dinger's cat: Does a file exist before you open it?\n\nRegister here\nWe are waiting for you!
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 2
UID:eventx6a5a287eec5ed10716
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20250113T173000
DTEND;TZID=Asia/Jerusalem:20250113T193000
DTSTAMP;TZID=Asia/Jerusalem:20250113T173000
SUMMARY: CSpecial Event  The Complete Guide to Industrial Jobs - The Workshop That Will Get You Organized!  at 2025-01-13 17:30:00
DESCRIPTION:It's time to take your career a step further!\nA career marathon is underway, and you are invited to our first meeting - an exposure workshop for jobs in the industry - taking stock at Buzz Words -Monday, 13.1, at 17:30 in the Piano Auditorium\nWant to understand what's behind the coveted titles in the industry?In this workshop, we will take stock of the variety of job types, dive into innovative technologies and products, and bring you a real glimpse into a typical workday in leading positions.\nWhat's on the program:\n17:30 Short, focused lectures on the areas:\nDevOps: Eyal Grover, Senior Software and DevOps Manager at NVIDIA on the challenges and tools that shape the world of DevOps.\nAI and Machine Learning: Dr. Eli Schwartz, Senior Researcher at IBM on groundbreaking developments in machine learning and their applications in industry.\nEmbedded Systems: Tawfik Boyk, Senior Director at Mobileye on the development of embedded systems and their place in the world of autonomous vehicles.\n18:15 Q&amp;A panel with the experts:An opportunity to delve deeper, ask questions, and hear firsthand from leading experts in their fields!\nThis is your opportunity to receive tools that will prepare you for the next step in your career.\nRegister here (limited spaces)\nWe are waiting for you!
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Piano Auditorium (012)
UID:eventx6a5a287eec5ff10721
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20250114T103000
DTEND;TZID=Asia/Jerusalem:20250114T123000
DTSTAMP;TZID=Asia/Jerusalem:20250114T103000
SUMMARY: colloq  talk by Noam Mazor (Tel Aviv University)  Computational Analogs of Randomness  at 2025-01-14 10:30:00
DESCRIPTION:Computational analogs of information-theoretic notions have given rise to some of the most intriguing phenomena in theoretical computer science. For example, pseudorandomness allows us to bypass Shannon's lower bounds on the key length of encryption schemes. Moreover, computational analogs of entropy and randomness are key tools in the construction of pseudorandom generators and have become foundational concepts in complexity theory and cryptography.One such computational analog is time-bounded Kolmogorov complexity. This measure lies at the heart of the emerging connections between cryptography and Kolmogorov complexity. Despite its significance, fundamental questions about the hardness of computing this measure remain open. In this talk, we will explore these questions together with recent advancements. We will also discuss how understanding the computational hardness of meta-complexity problems is instrumental in characterizing the existence of cryptographic primitives.&nbsp;\nBio: Noam completed his PhD at Tel Aviv University under the supervision of Iftach Haitner. For his PhD work, he received the Blavatnik Prize for Outstanding Israeli PhD Students in Computer Science and the Best Young Researcher Award at TCC 2021. During his PhD, he also received a teaching excellence award at Tel Aviv University. Since then, Noam has been a postdoctoral fellow at Cornell Tech and Tel Aviv University, hosted by Rafael Pass and Joseph Halpern, and in 2025, he will join the Cryptography Semester at the Simons Institute as a research fellow. His research focuses primarily on the foundations of cryptography, privacy, and meta-complexity.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Room 337
UID:eventx6a5a287eec61310724
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20250114T113000
DTEND;TZID=Asia/Jerusalem:20250114T133000
DTSTAMP;TZID=Asia/Jerusalem:20250114T113000
SUMMARY: pixel-club  talk by Yoni Kasten (NVIDIA)  Pixel Club - Algebraic Approaches and Deep Neural Models for 3D Scene Reconstruction and Camera Pose Estimation in Static and Dynamic Environments  at 2025-01-14 11:30:00
DESCRIPTION:Tuesday, January 14, 2025 at 11:30Meyer Building Room 1061 &amp; Zoom\nThis talk will explore advances in 3D scene reconstruction, focusing on approaches to estimate camera poses and scene structures in challenging multiview and dynamic content scenarios. First, I will outline foundational aspects of my earlier work, where we characterized the algebraic structure of fundamental and essential matrices in multiview settings and developed deep learning methods for joint recovery of camera parameters and sparse 3D scene structures. The main part of the talk introduces TracksTo4D (NeurIPS 2024), a novel, efficient method for reconstructing dynamic 3D structures and camera motion from casual videos. TracksTo4D leverages a dedicated encoder, trained in an unsupervised way on a dataset of casual videos, that uses 2D point tracks as input to infer dynamic 3D structures and camera motion. Our architecture takes into account symmetries in the problem, enforces the reconstruction to be of low rank, and models both static and dynamic scene components. Our model demonstrates strong generalization to unseen videos from new categories, achieving accurate 3D reconstruction and camera localization through a single feed-forward pass while drastically reducing running times.&nbsp;Short bio: Yoni Kasten is a senior research scientist at NVIDIA Research in Tel Aviv, in Prof. Gal Chechik&rsquo;s team. His research in 3D computer vision focuses on algebraic characterizations of multi-camera systems and deep neural models for surface reconstruction, dynamic scene modeling, and 4D scene reconstruction. Yoni earned his PhD from the Weizmann Institute, where his work on structure from motion estimation using algebraic characterizations, supervised by Prof. Ronen Basri, received the John F. Kennedy Prize for Outstanding Doctoral Research. He also completed his M.Sc. in Computer Science at the Hebrew University of Jerusalem, under the supervision of Prof. Shmuel Peleg and Prof. Michael Werman.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:
UID:eventx6a5a287eec62610723
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20250115T123000
DTEND;TZID=Asia/Jerusalem:20250115T133000
DTSTAMP;TZID=Asia/Jerusalem:20250115T123000
SUMMARY: MSC  talk by Asaf Levi  Indexing Deduplicated Storage  at 2025-01-15 12:30:00
DESCRIPTION:Deduplication is widely utilized in many modern large scale storage systems and provide an effective solution for both secondary and primary storage. Therefore, there is a rising need for deduplication storage to support advanced features such as data indexing for information retrieval. To our knowledge, no indexing solution for deduplicated storage utilizes the deduplication and current indexing methods process duplicates.\nIn this work, we propose IDEA, Inverted Deduplication-Aware Index, which we use to explore the potential of utilizing deduplication in keyword-indexing. IDEA is shown to be superior to the deduplication-oblivious approach, in both index creation and index size, and index query retrieval time. IDEA is also shown to be extendible for advanced indexing features, and orthogonal to the underlying index-engine.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 8 &amp; Zoom: 92977973231
UID:eventx6a5a287eec63b10702
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20250115T123000
DTEND;TZID=Asia/Jerusalem:20250115T143000
DTSTAMP;TZID=Asia/Jerusalem:20250115T123000
SUMMARY: CSpecial Event  Final Company Spotlight Day January 15, 2025  at 2025-01-15 12:30:00
DESCRIPTION:Final Company is coming to the faculty for a spotlight day and technology lectureNext Wednesday, January 15 starting at 12:30 at Taub\nLooking for your next step? Want to hear about algo-trading, options and probabilities?Final Company is coming with recruitment teams and engineers who will tell you everything &ndash; from career to technology.\nWhat awaits you?&bull; 12:30 | Taub Lobby &ndash; Meeting with researchers, engineers and recruitment teams.&bull; 13:15 | Taub 9 &ndash; Lecture: &ldquo;The Right Price&rdquo; Options, Probabilities and the World of Algo-TradingSpeaker-Noam Horvitz, researcher at Final.\nAnd... yes, there are refreshmentsRegister at the link here\nWe are waiting for you!!
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub Lobby / Taub 9
UID:eventx6a5a287eec64b10722
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20250115T130000
DTEND;TZID=Asia/Jerusalem:20250115T140000
DTSTAMP;TZID=Asia/Jerusalem:20250115T130000
SUMMARY: Theory Semina  talk by Dor Minzer (MIT)  Theory Seminar: On Approximability of Satisfiable CSPs and Friends  at 2025-01-15 13:00:00
DESCRIPTION:Constraint satisfaction problems (CSPs in short) are among the most important computational problems studied in TCS.&nbsp;This talk will focus on a recent line of study addressing the complexity of approximating satisfiable instances of CSPs, and&nbsp; connections of this study to multi-player parallel repetition theorems, property testing and extremal combinatorics.\nBased mostly on joint works with Amey Bhangale, Subhash Khot and Yang P. Liu.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 4
UID:eventx6a5a287eec65e10727
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20250116T120000
DTEND;TZID=Asia/Jerusalem:20250116T130000
DTSTAMP;TZID=Asia/Jerusalem:20250116T120000
SUMMARY: MSC  talk by Yonatan Sommer  Learning Classifiers That Induce Markets  at 2025-01-16 12:00:00
DESCRIPTION:When learning is used to inform decisions about humans, such as for loans, hiring, or admissions, this can incentivize users to strategically modify their features to obtain positive predictions. A key assumption is that modifications are costly, and&nbsp;are governed by a cost function that is exogenous and predetermined. We challenge this assumption, and assert that the deployment of a classifier is what creates costs. Our idea is simple: when users seek positive predictions, this creates demand for important features; and if features are available for purchase, then a market will form, and competition will give rise to prices. We extend the strategic classification framework to support this notion, and study learning in a setting where a classifier can induce a market for features. We present an analysis of the learning task, devise an algorithm for computing market prices, propose a differentiable&nbsp;learning framework, and conduct experiments&nbsp;to explore our novel setting and approach.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 8 &amp; Zoom
UID:eventx6a5a287eec67010725
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20250121T103000
DTEND;TZID=Asia/Jerusalem:20250121T123000
DTSTAMP;TZID=Asia/Jerusalem:20250121T103000
SUMMARY: colloq  talk by Yossi Gandelsman (UC Berkeley)  Interpreting the Inner Workings of Vision Models  at 2025-01-21 10:30:00
DESCRIPTION:In this talk, I present an approach for interpreting the internal computation in deep vision models. I show that these interpretations can be used to detect model bugs and to improve the performance of pre-trained deep neural networks (e.g., reducing hallucinations from image captioners and detecting and removing spurious correlations in CLIP) without any additional training. Moreover, the obtained understanding of deep representations can unlock new model capabilities (e.g., novel identity editing techniques in diffusion models and faithful image inversion in GANs). I demonstrate how to find common representations across different models (discriminative and generative) and how deep representations can be adapted at test-time to improve model generalization without any additional supervision. Finally, I discuss future work on improving the presented interpretation techniques and their application to continual model correction and scientific discovery.\nBio:Yossi is a EECS PhD at UC Berkeley, advised by Alexei Efros, and a visiting researcher at Meta. Before that, he was a member of the perception team at Google Research (now Google-DeepMind). He completed his M.Sc. at Weizmann Institute, advised by Prof. Michal Irani. His research centers around deep learning, computer vision, and mechanistic interpretability.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Room 337
UID:eventx6a5a287eec68110729
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20250121T113000
DTEND;TZID=Asia/Jerusalem:20250121T133000
DTSTAMP;TZID=Asia/Jerusalem:20250121T113000
SUMMARY: pixel-club  talk by Or Patashnik (Tel Aviv University)  Pixel Club - Leveraging Pretrained Generative Models for Real Image Editing  at 2025-01-21 11:30:00
DESCRIPTION:Image generative models are advancing rapidly, producing images of remarkable realism and fidelity. However, existing models often lack precise control over the generated content, limiting their image editing capabilities and the integration of real content into synthesized imagery. In this talk, I will demonstrate how a deep understanding of the inner mechanisms of large-scale pretrained generative models enables the design of powerful techniques for a variety of image manipulation tasks. By analyzing the semantic representations learned by these models, I will present methods that enable effective content editing. Additionally, I will discuss the challenges and trade-offs involved in manipulating real content and propose strategies to address these challenges. Finally, I will highlight recent advancements in incorporating real content, with a particular focus on techniques for injecting information into pretrained models.\nBio:&nbsp;Or Patashnik (https://orpatashnik.github.io/) is a Computer Science PhD candidate at Tel Aviv University, supervised by Daniel Cohen-Or. Her research focuses on computer graphics and its intersection with computer vision, with an emphasis on generative tasks such as image editing, personalization, and image inversion using large-scale pretrained models. Recently, she has been particularly interested in better understanding diffusion models for various applications.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Room 1061 Meyer Building &amp; Zoom
UID:eventx6a5a287eec69610731
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20250121T173000
DTEND;TZID=Asia/Jerusalem:20250121T193000
DTSTAMP;TZID=Asia/Jerusalem:20250121T173000
SUMMARY: CSpecial Event  My First Resume Workshop - 21.1.25  at 2025-01-21 17:30:00
DESCRIPTION:Continuing the career workshop marathon!\nAfter you have become familiar with the types of jobs available, it is time to upgrade your resume.\nWe invite you to the My First Resume Workshop - How to Turn a Blank Page into an Opportunity, hosted by Bar Yaakovi, a graduate of the FacultyTuesday, 21.1, starting at 5:30 PM at Taub 337\nAbout the speaker:Bar Yaakovi holds a bachelor's degree in computer science from the Technion and is currently a master's student at Tel Aviv University. In her role as a software engineer at Meta and other leading companies in the industry, she has gained extensive experience in analyzing and reviewing resumes. She has often encountered resumes from talented candidates who fail to present themselves in the best possible way.\nFeel like you have nothing to write? In the workshop, Bar will share professional insights that will help you identify your strengths, avoid common mistakes, and build an impressive resume that will make you stand out in the job market.\nRegister here\nDon't miss the opportunity to take the first step towards a successful career!
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 337
UID:eventx6a5a287eec6d110734
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20250122T103000
DTEND;TZID=Asia/Jerusalem:20250122T123000
DTSTAMP;TZID=Asia/Jerusalem:20250122T103000
SUMMARY: colloq  talk by Ofir Press (Princeton University)  Towards Autonomous Language Model Systems  at 2025-01-22 10:30:00
DESCRIPTION:Language models (LMs) are increasingly used to assist users in day to day tasks such as programming (Github Copilot) or search (Google's AI Overviews). But can we build language model systems that are able to autonomously complete entire tasks end-to-end? In this talk I'll discuss our efforts to build autonomous LM systems, focusing on the software engineering domain. I'll present SWE-bench, our novel method for measuring the performance of automatic programming systems on their abilities to fix real issues in popular software libraries from GitHub. I'll then discuss SWE-agent, our system for solving SWE-bench tasks. SWE-bench and SWE-agent are used by many leading AI orgs in academia and industry including OpenAI, Anthropic, Meta, and Google, and these projects show that academics on tight budgets are able to have substantial impact on steering the research community towards building autonomous systems that can complete challenging tasks.&nbsp;\nShort bio:Ofir Press is a postdoc at Princeton University. I previously completed my PhD at the University of Washington in Seattle, where I was advised by Noah Smith. During my PhD I spent two years at Facebook AI Research Labs.&nbsp;
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 337
UID:eventx6a5a287eec6f710735
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20250122T110000
DTEND;TZID=Asia/Jerusalem:20250122T120000
DTSTAMP;TZID=Asia/Jerusalem:20250122T110000
SUMMARY: MSC  talk by Tal Neoran  Histopathology Whole Slide Image Analysis by Weakly Supervised and Self-Supervised Deep Learning  at 2025-01-22 11:00:00
DESCRIPTION:Digital pathology has emerged as a transformative field, enabling automated imaging and computational analysis of thin tissue biopsy slices or body fluids. These samples are typically stained to enhance contrast in biological structures and reveal their morphology under microscopic magnification. Digitally scanning these stained samples produces gigapixel-scale images, known as whole slide images (WSIs), which pose significant challenges for computational analysis using deep learning techniques. Variability in staining protocols and scanning devices across medical centers introduces inconsistencies in WSIs, while their large size imposes substantial computational constraints. Furthermore, most slides are annotated such that the pathologist's prognosis is provided as a holistic textual report at the patient level, rather than pinpointing the specific diagnostic regions in the image that indicate the medical estimate.\nTo address these challenges, we propose an end-to-end approach for WSI analysis. Our method involves self-supervised pretraining on patches extracted from a large collection of WSIs to learn generic and transferable feature representations without requiring manual annotations. Subsequently, we employ a Transformer-based architecture trained in a weakly supervised manner to effectively aggregate spatial and contextual information across image regions, producing accurate slide-level predictions.\nWe evaluated our proposed method on multiple WSI datasets for various clinically relevant molecular profiling tasks, such as determining hormonal receptor status in invasive breast cancer, demonstrating its effectiveness and generalizability. Leveraging recent advances in computer vision, our approach addresses the challenges of WSI analysis, providing insights for processing large and complex images. By enhancing robustness and predictive accuracy in various clinical tasks within computational histopathology, our approach has the potential to improve clinical decision-making, ultimately contributing to better patient outcomes.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 401 &amp; Zoom
UID:eventx6a5a287eec70c10733
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20250122T130000
DTEND;TZID=Asia/Jerusalem:20250122T140000
DTSTAMP;TZID=Asia/Jerusalem:20250122T130000
SUMMARY: Theory Semina  talk by Noga Amit (UC Berkeley)  Theory Seminar: Models that prove their own correctness  at 2025-01-22 13:00:00
DESCRIPTION:How can we trust the correctness of a learned model on a particular input of interest? Model accuracy is typically measured on average over a distribution of inputs, giving no guarantee for any specific input. This talk introduces Self-Proving models, a new class of models that formally prove the correctness of their outputs via an Interactive Proof system. We will formally define Self-Proving models and their per-input (worst-case) guarantees. We will then present algorithms for learning these models and explain how the complexity of the proof system affects the complexity of the learning algorithms. Finally, we will review experiments in which Self-Proving Models are trained to compute the greatest common divisor (GCD) of two integers and prove their correctness to a simple verifier.\nNo prior knowledge of autoregressive models or Interactive Proofs will be assumed from the audience. This is a joint work with Shafi Goldwasser, Orr Paradise and Guy Rothblum.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 4
UID:eventx6a5a287eec72210737
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20250122T170000
DTEND;TZID=Asia/Jerusalem:20250122T200000
DTSTAMP;TZID=Asia/Jerusalem:20250122T170000
SUMMARY: CSpecial Event  Welcome to the "AI on the go: Programming the AI-PC" workshop - On behalf of Intel  at 2025-01-22 17:00:00
DESCRIPTION:The AI ​​revolution is expanding from the cloud directly to our personal computers, and you are invited to join a 3-hour hands-on workshop where you can experience the latest advances and capabilities.\nDuring the workshop, you will delve into the new technologies of Intel&rsquo;s AI-PCs, build chatbots, produce photos and videos, and discover how to create music on your laptop.\nThe workshop is intended for anyone interested in expanding their knowledge in the field of AI and requires basic knowledge of Python.\nThere is space for the first 50 registrants, please fill in the details to register and secure your place!\nThe workshop will take place on January 22nd between 5:00 PM and 8:00 PM in Taub 9 room\nRegister here - hurry up and register, the number of places is limited!
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 9
UID:eventx6a5a287eec73410704
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20250127T153000
DTEND;TZID=Asia/Jerusalem:20250127T163000
DTSTAMP;TZID=Asia/Jerusalem:20250127T153000
SUMMARY: MSC  talk by Tomer Gewirtzman  Zero-Knowledge in Streaming Interactive Proofs  at 2025-01-27 15:30:00
DESCRIPTION:In a recent work, Cormode, Dall'Agnol, Gur and Hickey (CCC, 2024) introduced the model of Zero-Knowledge&nbsp;Streaming Interactive Proofs (zkSIPs).&nbsp;Loosely speaking, such proof-systems enable a prover to convince a&nbsp;streaming verifier that the input x, to which it has read-once streaming access, satisfies some property, in&nbsp;such a way that nothing beyond the correctness of the claim is revealed.&nbsp;Cormode et al. also gave constructions of zkSIPs to some specific and notable problems of interest.\nIn this work, we advance the study of zero-knowledge proofs in the streaming model, by presenting protocols that&nbsp;are significantly more general and more secure.&nbsp;We use a definition of zero-knowledge that is a variation of that used by Cormode et al., which we find more&nbsp;appealing but is technically incomparable.\nOur main result is a zkSIP for any NP relation, that can be decided by low-depth polynomial-size circuits.&nbsp;We emphasize that this is the first general purpose protocol in this model, which captures, as a special&nbsp;case, the problems considered by the prior work.&nbsp;We also construct a specialized protocol for the ``polynomial evaluation'' problem considered in that work, with&nbsp;improved parameters.\nThe protocols constructed by Cormode et al. have an inverse polylogarithmic simulation error (i.e., a gap with&nbsp;which a bounded-space distingiusher can distinguish the simulation from a real execution).&nbsp;This means that their protocols are entirely insecure if run multiple times (say on different inputs).&nbsp;In contrast, our protocols achieve a negligible zero-knowledge error, a stronger and far more robust security&nbsp;guarantee.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Room 601 &amp; Zoom
UID:eventx6a5a287eec74710726
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20250128T103000
DTEND;TZID=Asia/Jerusalem:20250128T233000
DTSTAMP;TZID=Asia/Jerusalem:20250128T103000
SUMMARY: colloq  talk by Or Patashnik (Tel Aviv University)  Leveraging Pretrained Generative Models for Real Image Editing  at 2025-01-28 10:30:00
DESCRIPTION:Image generative models are advancing rapidly, producing images of remarkable realism and fidelity. However, existing models often lack precise control over the generated content, limiting their image editing capabilities and the integration of real content into synthesized imagery. In this talk, I will demonstrate how a deep understanding of the inner mechanisms of large-scale pretrained generative models enables the design of powerful techniques for a variety of image manipulation tasks. By analyzing the semantic representations learned by these models, I will present methods that enable effective content editing. Additionally, I will discuss the challenges and trade-offs involved in manipulating real content and propose strategies to address these challenges. Finally, I will highlight recent advancements in incorporating real content, with a particular focus on techniques for injecting information into pretrained models. &nbsp;\nBio: Or Patashnik (https://orpatashnik.github.io/) is a Computer Science PhD candidate at Tel Aviv University, supervised by Daniel Cohen-Or. Her research focuses on computer graphics and its intersection with computer vision, with an emphasis on generative tasks such as image editing, personalization, and image inversion using large-scale pretrained models. Recently, she has been particularly interested in better understanding diffusion models for various applications.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 337
UID:eventx6a5a287eec77c10740
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20250129T103000
DTEND;TZID=Asia/Jerusalem:20250129T123000
DTSTAMP;TZID=Asia/Jerusalem:20250129T103000
SUMMARY: colloq  talk by Oded Stein (USC’s Viterbi School of Engineering)  Doing More With Less in Geometry Processing  at 2025-01-29 10:30:00
DESCRIPTION:Geometric data and signal processing have made remarkable progress in recent years, and the advent of modern AI tools promises an even brighter future. Many classical and contemporary methods, however, use vast amounts of data, substantial processing power, and considerable natural resources to achieve their results. These approaches can be expensive, environmentally harmful, and ultimately unsustainable. This talk explores efforts to do more with less in geometry processing by ensuring that methods use all the information contained in input data, by maximally utilizing noisy and sparse data, and by developing methods for computer-aided fabrication with sustainable materials. We will dive deeper into surface digitization from extremely low-resolution scans, training generative AI models without collecting terabytes of data, denoising functions on low-quality geometric domains, developing robust geometric algorithms with mathematical guarantees, and design tools for alternative manufacturing methods.\nBio: Oded Stein received his PhD from Columbia University in Applied Mathematics from Columbia University under the supervision of Eitan Grinspun for his research of smoothness energies in geometry processing in 2020. From 2020 to 2022 he was a postdoctoral fellow at MIT&rsquo;s Computer Science &amp; Artificial Intelligence Lab with the gracious support of the Swiss National Science Foundation&rsquo;s Early Postdoc. Mobility fellowship. Since 2023 Oded Stein is an Assistant Professor at USC&rsquo;s Viterbi School of Engineering, supported by the Powell Faculty Research Award and the National Science Foundation.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Room 337
UID:eventx6a5a287eec78e10741
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20250129T110000
DTEND;TZID=Asia/Jerusalem:20250129T120000
DTSTAMP;TZID=Asia/Jerusalem:20250129T110000
SUMMARY: PHD  talk by Omer Sabary  Coding Solutions and Algorithms for Emerging Synthesis and Sequencing Technologies  at 2025-01-29 11:00:00
DESCRIPTION:Over the past decade, several studies have shown that DNA-based storage systems can potentially become the standard for data archival due to their high data density and durability. However, the current bottleneck involves the synthesis and sequencing costs, along with a lack of coding solutions to address the unique error characteristics of DNA-based systems.\nThis work tackles multiple challenges that hinder the practical implementation of DNA storage. First, we explore theoretical aspects of the deletion channel, presenting detailed findings from the maximum likelihood decoder for both single and dual-channel outputs. Next, we address the DNA reconstruction problem, aiming to accurately reconstruct a DNA sequence from multiple noisy copies. We propose several reconstruction algorithms that significantly enhance accuracy compared to previously published approaches. Furthermore, we investigate two novel synthesis methods, the composite synthesis and the combinatorial composite synthesis, highlighting their potential benefits and inherent complexities. These methods require innovative algorithmic and coding solutions, and thus we design error-correction codes specifically tailored for these technologies.\nFinally, we introduce the DNA storalator, a software tool designed to simulate the biological and computational processes of DNA storage, aiding our research and facilitating further exploration within the scientific community. Overall, the results presented in this work advance several aspects of DNA data storage and promote the feasibility of this storage solution further.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 601 &amp; Zoom&nbsp;
UID:eventx6a5a287eec7a810728
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20250129T113000
DTEND;TZID=Asia/Jerusalem:20250129T123000
DTSTAMP;TZID=Asia/Jerusalem:20250129T113000
SUMMARY: ceClub  talk by Dr. Ben Nassi (Technion)  CE-Club - Securing Modern Systems is More Challenging Than Ever (and Requires New and Dedicated Guardrails)  at 2025-01-29 11:30:00
DESCRIPTION:&nbsp;Over the past decade, an increasing number of systems and devices have gained Internet connectivity and been enhanced with sensing capabilities and AI. While these advancements have created a world of smarter, more automated, and highly connected devices, they have also introduced significant security and privacy challenges that cannot be effectively addressed with traditional countermeasures.\nIn the first part of this talk, we will explore the security and privacy concerns of cyber-physical systems. Specifically, we will examine new threats that have emerged with the deployment of technologies like drones and Teslas in real-world environments. Our discussion will highlight methods for detecting intrusive drone filming and securing Teslas against time-domain adversarial attacks.\nThe second part of the talk focuses on the challenges posed by the coexistence of functional devices with limited computational power (that do not adhere to Moore&rsquo;s law) alongside sensors with ever-increasing sampling rates. We will explore how threats such as cryptanalysis and speech eavesdropping&mdash;previously accessible only to well-resourced adversaries&mdash;can now be executed by ordinary attackers using readily available hardware like photodiodes and video cameras. These attacks leverage optical traces or video footage from a device&rsquo;s power LED to extract sensitive information.\nFinally, in the last part of the talk, we will address the emerging need to secure GenAI-powered applications against a new category of threats we call Promptware. This threat highlights the evolving landscape of vulnerabilities introduced by generative AI systems.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zisapel Building 506
UID:eventx6a5a287eec7c510736
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20250129T123000
DTEND;TZID=Asia/Jerusalem:20250129T143000
DTSTAMP;TZID=Asia/Jerusalem:20250129T123000
SUMMARY: CSpecial Event  A Dedicated Event For Graduate Students With StarkWare 29.1.25  at 2025-01-29 12:30:00
DESCRIPTION:The Faculty of Computer Science invites you to a dedicated event for graduate students of StarkWare, a leader in blockchain technologies - "Engineering the Future of Blockchain"\nThis coming Wednesday, January 29, 25, at 12:30, in the Piano Auditorium (note the change in the lecture location).\nWhat's on the program:12:30 How StarkWare is leading the blockchain revolution with innovative solutions13:30 Applying algorithms to real-world solutions14:00 Mingling and refreshments\nRegister here\nDon't miss the opportunity to get to know the technologies of the future!
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Piano Auditorium
UID:eventx6a5a287eec7fe10746
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20250129T130000
DTEND;TZID=Asia/Jerusalem:20250129T140000
DTSTAMP;TZID=Asia/Jerusalem:20250129T130000
SUMMARY: Theory Semina  talk by Or Zamir (Tel-Aviv University)  Theory Seminar: Optimality of Frequency Moment Estimation  at 2025-01-29 13:00:00
DESCRIPTION:Estimating the second frequency moment of a stream up to (1 &plusmn; &epsilon;) multiplicative error requires at most O(log n / &epsilon;&sup2;) bits of space, due to a seminal result of Alon, Matias, and Szegedy. It is also known that at least &Omega;(log n + 1/&epsilon;&sup2;) space is needed.\nWe prove a tight lower bound of &Omega;(log(n&epsilon;&sup2;) / &epsilon;&sup2;) for all &epsilon; = &Omega;(1/&radic;n).\nNotably, when &epsilon; &gt; n^(-1/2 + c), where c &gt; 0, our lower bound matches the classic upper bound of AMS. For smaller values of &epsilon;, we also introduce a revised algorithm that improves the classic AMS bound and matches our lower bound.\nOur lower bound also applies to the more general problem of p-th frequency moment estimation for the range of p in (1, 2], providing a tight bound in the only remaining range to settle the optimal space complexity of estimating frequency moments.\nBased on a joint work with Mark Braverman.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 4
UID:eventx6a5a287eec81010743
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20250129T170000
DTEND;TZID=Asia/Jerusalem:20250129T180000
DTSTAMP;TZID=Asia/Jerusalem:20250129T170000
SUMMARY: PHD  talk by Ilan Doron-Arad  Tight Bounds for Matroid Problems with a Linear Constraint  at 2025-01-29 17:00:00
DESCRIPTION:We study budgeted variants of the well known Matching, Matroid Independent Set, and Matroid Intersection problems. While the three problems admit polynomial-time approximation schemes (PTAS) [Berger et al. (Math. Programming, 2011), Chekuri, Vondrak and Zenklusen (SODA 2011)], it has been an intriguing open question whether these problems admit a Fully PTAS (FPTAS), or even an Efficient PTAS (EPTAS). In this work, we show&nbsp;that the three problems admit an EPTAS. On the other hand, we rule out the existence of an FPTAS for Budgeted Matroid Independent Set and Budgeted Matroid Intersection, thus resolving the complexity status of the last two problems.\nWe extend our lower bounds to the more general family of matroid optimization problems with a linear constraint (MOL). In these problems, we seek a subset of elements which optimizes (i.e., maximizes or minimizes) a linear objective function subject to (i) a matroid independent set, or a matroid basis constraint, (ii) an additional linear constraint. We show that none of the (non-trivial) problems in this family admits a Fully PTAS. This resolves the complexity status of several well studied problems, including Constrained Minimum Basis of a Matroid and Knapsack Cover with a Matroid Constraint.\nOur EPTASs rely on the novel representative set technique in which we replace exhaustive enumeration over a given input by enumeration over a relatively small subset of elements. We obtain our lower bounds for MOL problems by showing first that the classic Exact Weight Matroid Basis (EMB) problem does not admit a pseudo-polynomial time algorithm. This distinguishes EMB from the special cases of k-subset sum and EMB on a linear matroid, which are solvable in pseudo-polynomial time.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 8 &amp; Zoom
UID:eventx6a5a287eec82310742
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20250205T090000
DTEND;TZID=Asia/Jerusalem:20250205T163000
DTSTAMP;TZID=Asia/Jerusalem:20250205T090000
SUMMARY: CSpecial Event  AI & BEYOND Robotics Conference February 5, 2025  at 2025-02-05 09:00:00
DESCRIPTION:We are happy to invite you to the Tech.AI Robotics Conference, happening on February 5, 2025. This exciting event will explore the critical crossroads of artificial intelligence and robotics: from smart home systems and autonomous vehicles to advanced manufacturing and medical procedures. These groundbreaking technologies are revolutionizing research and redefining how we live and work.\nJoin us for a dynamic program filled with diverse and insightful talks delivered by leading experts.\nFor more information and to secure your tickets click here. If you have any questions or need further assistance, don&rsquo;t hesitate to reach out to us at Tech-ai@technion.ac.il.\nWe can&rsquo;t wait to see you at the conference!
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Electrical and Computer Engineering Faculty
UID:eventx6a5a287eec83a10719
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20250205T113000
DTEND;TZID=Asia/Jerusalem:20250205T123000
DTSTAMP;TZID=Asia/Jerusalem:20250205T113000
SUMMARY: PHD  talk by Oleg Kolosov  Workloads, Storage, and Service Allocation in Edge Computing  at 2025-02-05 11:30:00
DESCRIPTION:Edge computing extends cloud capabilities to the proximity of end-users, offering ultra-low latency, which is essential for real-time applications. Unlike traditional cloud systems that suffer from latency and reliability constraints due to distant datacenters, edge computing employs a distributed model, leveraging local edge datacenters to process and store data.\nThis talk explores key challenges in edge computing across three domains: workloads, storage, and service allocation.&nbsp;The first part focuses on the absence of comprehensive edge workload datasets. Current datasets do not accurately reflect the unique attributes of edge systems. To address this, we propose a workload composition methodology and introduce WoW-IO, an open-source trace generator. The second part examines aspects of edge storage. Edge datacenters are significantly smaller than their cloud counterparts and require dedicated solutions. We analyze the applicability of a promising mathematical model for edge storage systems and raise inherent gaps between theory and practice. The final part addresses the virtual network embedding problem (VNEP). In VNEP, given a set of requests for deploying virtualized applications, the edge provider has to deploy a maximum number of them to the underlying physical network, subject to capacity constraints. We propose novel solutions, including a proactive service allocation strategy for mobile users, a flexible algorithm for service allocation that is adaptable to the underlying physical topology, and &nbsp;an algorithm for scalable online service allocation.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 8 &amp; Zoom
UID:eventx6a5a287eec85110730
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20250210T110000
DTEND;TZID=Asia/Jerusalem:20250210T120000
DTSTAMP;TZID=Asia/Jerusalem:20250210T110000
SUMMARY: MSC  talk by Andrew Luka  Property Directed Reachability with Extended Resolution  at 2025-02-10 11:00:00
DESCRIPTION:Property Directed Reachability (Pdr), also known as IC3, is a state-of-the-art model checking algorithm widely used for verifying safety properties. While Pdr is effective in finding inductive invariants, its underlying proof system, Resolution, limits its ability to construct short proofs for certain verification problems.\nIn this talk we present PdrER, a generalization of Pdr that uses Extended Resolution (ER), a proof system exponentially stronger than Resolution. Using a strong proof system is not straight forward: the stronger a proof system is, the harder it is to implement an *efficient* proof-search algorithm for it. While strong proof systems, such as Extended Resolution, have been used in the context of Boolean Satisfiability (SAT), it has had very limited success.\nPdrER is the first model checking algorithm to use ER. This allows PdrER to construct safety proofs more efficiently, and to overall outperform Pdr on standard benchmarks. Achieving this result required careful engineering of the algorithm in order to overcome the overhead of the increased search-space associated with ER. We will delve into the details of this effort in this talk.\nWe implemented PdrER in a new open-source verification framework and evaluated it on the Hardware Model Checking Competition benchmarks from 2019, 2020 and 2024. Our experimental evaluation demonstrates that PdrER outperforms Pdr, solving more instances in less time and uniquely solving problems that Pdr cannot solve within a given time limit. We argue that this work represents a significant step toward making strong proof systems practically usable in model checking.\n&nbsp;
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 601 &amp; Zoom
UID:eventx6a5a287eec86510750
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20250210T153000
DTEND;TZID=Asia/Jerusalem:20250210T163000
DTSTAMP;TZID=Asia/Jerusalem:20250210T153000
SUMMARY: MSC  talk by Ron Alfia  Multimodal Robustness to Input Corruptions in 3D Object Detection  at 2025-02-10 15:30:00
DESCRIPTION:In the age of abundant data, deep learning has emerged as a leading tool for predictive tasks, consistently setting new benchmarks in areas such as computer vision. One such task is 3D Object Detection (3DOD), where the goal is to estimate the locations of objects within a 3D space using inputs like RGB images and LiDAR point clouds. This task is crucial for applications in advanced driver-assistance systems, autonomous vehicles, and robotic navigation. Despite the promise of deep learning, these models are known to often learn shallow correlations, a challenge particularly evident in multimodal setups. Specifically, when multimodal 3DOD models rely on both images and point clouds, they tend to exhibit a significant bias towards the point cloud data, even though both modalities contain overlapping information. This modality bias renders the model vulnerable to failures in LiDAR data, a crucial problem in real-world environments.\nIn this work, we define and quantify this modality bias and propose two novel debiasing techniques for multimodal 3DOD models. Additionally, we introduce a failure estimator task for point clouds and demonstrate its high accuracy using a novel architecture. Finally, we present an inference-only architecture that combines debiasing and failure estimation, resulting in a robust end-to-end multimodal model. This model not only performs well under normal conditions but also significantly improves accuracy in the presence of LiDAR failures, addressing a critical gap in robustness for multimodal 3DOD systems.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 601 &amp; Zoom
UID:eventx6a5a287eec87910745
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20250211T110000
DTEND;TZID=Asia/Jerusalem:20250211T120000
DTSTAMP;TZID=Asia/Jerusalem:20250211T110000
SUMMARY: MSC  talk by Einav Huberman  Near-Optimal Resilient Labeling Schemes  at 2025-02-11 11:00:00
DESCRIPTION:Labeling schemes are a prevalent paradigm in various computing settings. In such schemes, an oracle is given an input graph and produces a label for each of its nodes, enabling the labels to be used for various tasks. In this talk, I will address the question of what happens in a labeling scheme if some labels are erased, e.g., due to communication loss with the oracle or hardware errors. I will present a new resilient labeling scheme which improves upon the state of the art in several computational aspects and I will show that it is nearly optimal.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 8 &amp; Zoom
UID:eventx6a5a287eec88c10732
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20250211T140000
DTEND;TZID=Asia/Jerusalem:20250211T150000
DTSTAMP;TZID=Asia/Jerusalem:20250211T140000
SUMMARY: MSC  talk by Yehonatan Lusky  Reverse Engineering based Extraction of Convolutional Neural Networks  at 2025-02-11 14:00:00
DESCRIPTION:The extraction of neural networks poses a significant challenge to the security and intellectual property of AI models, enabling adversaries to recreate proprietary architectures, breach confidentiality, and exploit model functionality. In this seminar talk, I will introduce a novel attack that reconstructs both the structure and exact parameters of black-box convolutional neural networks (CNNs), using only query-based access. This technique is the first to recover the precise weight values and architecture of black-box CNNs. This method allows the extraction of common CNN models, including LeNet-5, AlexNet, and various VGG and ResNet architectures. I will outline the theoretical foundations of the attack and demonstrate its effectiveness through extractions of multiple architectures. This work highlights the real-world feasibility of model extraction and its broader implications for AI security.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 601 &amp; Zoom
UID:eventx6a5a287eec89b10748
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20250212T113000
DTEND;TZID=Asia/Jerusalem:20250212T123000
DTSTAMP;TZID=Asia/Jerusalem:20250212T113000
SUMMARY: MSC  talk by Daniel Silver  Coordinate Flow for Implicit Neural Representation Video Compression  at 2025-02-12 11:30:00
DESCRIPTION:In the field of video compression, the pursuit for better quality at lower bit rates remains a long-lasting goal. Recent developments have demonstrated the potential of Implicit Neural Representation (INR) as a promising alternative to traditional transform-based methodologies. Video INRs can be roughly divided into frame-wise and pixel-wise methods according to the structure the network outputs. While the pixel-based methods are better for upsampling and parallelization, frame-wise methods demonstrated better performance. We introduce CoordFlow, a novel pixel-wise INR for video compression. It yields state-of-the-art results compared to other pixel-wise INRs and on-par performance compared to leading frame-wise techniques. The method is based on the separation of the visual information into visually consistent layers, each represented by a dedicated network that compensates for the layer's motion. When integrated, a byproduct is an unsupervised segmentation of video sequence. Objects motion trajectories are implicitly utilized to compensate for visual-temporal redundancies. Additionally, the proposed method provides inherent video upsampling, stabilization, inpainting, and denoising capabilities.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 601 &amp; Zoom
UID:eventx6a5a287eec8ab10738
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20250212T113000
DTEND;TZID=Asia/Jerusalem:20250212T123000
DTSTAMP;TZID=Asia/Jerusalem:20250212T113000
SUMMARY: MSC  talk by Marwa Mouallem   Asynchronous Authentication  at 2025-02-12 11:30:00
DESCRIPTION:A myriad of authentication mechanisms embody a continuous evolution from verbal passwords in ancient times to contemporary multi-factor authentication: Cryptocurrency wallets advanced from a single signing key to using a handful of well-kept credentials, and for online services, the infamous &ldquo;security questions&rdquo; were all but abandoned. Nevertheless, digital asset heists and numerous identity theft cases illustrate the urgent need to revisit the fundamentals of user authentication. &nbsp;\nWe abstract away credential details and formalize the general, common case of asynchronous authentication, with unbounded message propagation time. Given credentials' fault probabilities (e.g., loss or leak), we seek mechanisms with maximal success probability. Such analysis was not possible before due to the large number of possible mechanisms. We show that every mechanism is dominated by some Boolean mechanism-defined by a monotonic Boolean function on presented credentials. &nbsp;We present an algorithm for finding approximately optimal mechanisms by leveraging the problem structure to reduce complexity by orders of magnitude. &nbsp;&nbsp;\nThe algorithm immediately revealed two surprising results: Accurately incorporating easily-lost credentials improves cryptocurrency wallet security by orders of magnitude. And novel usage of (easily-leaked) security questions improves authentication security for online services.&nbsp;
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Meyer 1061 (ECE building) &amp; Zoom
UID:eventx6a5a287eec8bc10739
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20250212T113000
DTEND;TZID=Asia/Jerusalem:20250212T133000
DTSTAMP;TZID=Asia/Jerusalem:20250212T113000
SUMMARY: ceClub  talk by Marwa Mouallem (Technion)  CE-Club - Asynchronous Authentication  at 2025-02-12 11:30:00
DESCRIPTION:A myriad of authentication mechanisms embody a continuous evolution from verbal passwords in ancient times to contemporary multi-factor authentication: Cryptocurrency wallets advanced from a single signing key to using a handful of well-kept credentials, and for online services, the infamous &ldquo;security questions&rdquo; were all but abandoned. Nevertheless, digital asset heists and numerous identity theft cases illustrate the urgent need to revisit the fundamentals of user authentication.&nbsp;\nWe abstract away credential details and formalize the general, common case of asynchronous authentication, with unbounded message propagation time. Given credentials' fault probabilities (e.g., loss or leak), we seek mechanisms with maximal success probability. Such analysis was not possible before due to the large number of possible mechanisms. We show that every mechanism is dominated by some Boolean mechanism-defined by a monotonic Boolean function on presented credentials. &nbsp;We present an algorithm for finding approximately optimal mechanisms by leveraging the problem structure to reduce complexity by orders of magnitude. &nbsp;&nbsp;\nThe algorithm immediately revealed two surprising results: Accurately incorporating easily-lost credentials improves cryptocurrency wallet security by orders of magnitude. And novel usage of (easily-leaked) security questions improves authentication security for online services.&nbsp;
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Meyer 1061 &amp; Zoom
UID:eventx6a5a287eec8cd10747
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20250212T140000
DTEND;TZID=Asia/Jerusalem:20250212T150000
DTSTAMP;TZID=Asia/Jerusalem:20250212T140000
SUMMARY: MSC  talk by Maayan Ehrenberg  Adversaries With Incentives: A Strategic Alternative to Adversarial Robustness  at 2025-02-12 14:00:00
DESCRIPTION:Adversarial training aims to defend against adversaries - malicious opponents aiming to harm predictive performance in any way possible. This strict perspective can result in overly conservative training. As an alternative, we propose modeling opponents as pursuing their own goals rather than working directly against the classifier. Employing tools from strategic modeling, our approach incorporates knowledge of the opponent's potential incentives as inductive bias for learning. We propose a method of strategic training designed to defend against all opponents within an `incentive uncertainty set'. This defaults to adversarial learning when the set is maximal, but offers potential gains when the set can be appropriately reduced.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 401
UID:eventx6a5a287eec8dd10744
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20250217T103000
DTEND;TZID=Asia/Jerusalem:20250217T113000
DTSTAMP;TZID=Asia/Jerusalem:20250217T103000
SUMMARY: MSC  talk by Alon Feldman  Planar Shape Interpolation: Bounding Conformal And Area Distortions Using Logarithmic Metric Blending  at 2025-02-17 10:30:00
DESCRIPTION:Shape interpolation is essential in graphics and geometry processing. For instance, smoothly transitioning between two poses of the same shape is crucial for animation, while morphing multiple shapes helps with design exploration. Since the blended shapes often differ, some distortion is unavoidable.\nWe introduce an interpolation method for planar shapes based on logarithmic metric blending. Our approach extends previous work on pullback metrics, enabling the use of various techniques - such as our proposed logarithmic blending - to precisely control both conformal and area distortions. A key contribution is the adaptation to discrete mesh interpolation through different conformal, isometric and equiareal parameterizations. Experimental results show that our method surpasses existing techniques in maintaining bounded distortions, making it an effective option for applications in animation and morphing.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom
UID:eventx6a5a287eec8eb10751
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20250217T123000
DTEND;TZID=Asia/Jerusalem:20250217T140000
DTSTAMP;TZID=Asia/Jerusalem:20250217T123000
SUMMARY: CSpecial Talk  by Michael Hanna  Mechanisms for formal and functional linguistic competence in LLMs  at 2025-02-17 12:30:00
DESCRIPTION:Abstract:&nbsp; As the capabilities of LLMs have grown, so has interest in using them to understand how language works in the human brain. Some have even suggested that LLMs do or should mimic how language is divided in the human brain: Mahowald et al. (2024)&nbsp;propose that, just as the brain has separate areas for formal linguistic&nbsp;competence (morphology, syntax, lexical semantics) as opposed to functional linguistic&nbsp;competence (world knowledge, reasoning, pragmatics), LLMs should learn a similar division&mdash;and may have even already done so. How can we translate such hypotheses from brains to LLMs, and test them in the latter? To answer this question, I'll first introduce circuits, a causal framework that allows us to localize the abilities of LLMs to a small set of components. Next, I'll apply these techniques to the question of formal and functional linguistic competence, showing that LLMs may in part learn distinct mechanisms for the two. Finally, I'll discuss alternative patterns&nbsp;that might underlie the mechanistic organization of LLMs.\nBio: Michael Hanna is a third-year PhD student at the University of Amsterdam, advised by Sandro Pezzelle and Yonatan Belinkov. His research lies at the intersection of mechanistic interpretability and cognitive science, using causal techniques to uncover the low-level mechanisms that support LLMs' strong&nbsp;linguistic abilities. To this end, he has worked on not only developing new methods for interpreting LLMs, but also applying these techniques to linguistic topics such as incremental sentence processing. Michael is an ELLIS PhD student and is supported by an OpenAI Superalignment Fellowship.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub, floor 0, Piano Auditorium
UID:eventx6a5a287eec8fb10752
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20250219T100000
DTEND;TZID=Asia/Jerusalem:20250219T110000
DTSTAMP;TZID=Asia/Jerusalem:20250219T100000
SUMMARY: cggc  talk by Prof. Mirela Ben-Chen  CGGC Seminar: Spectral Analysis of Coral Reef Deformation  at 2025-02-19 10:00:00
DESCRIPTION:We propose an efficient pipeline to register, detect, and analyze changes in 3D models of coral reefs captured over time. Corals have complex structures with intricate geometric features at multiple scales. 3D reconstructions of corals (e.g., using Pho- togrammetry) are represented by dense triangle meshes with millions of vertices. Hence, identifying correspondences quickly using conventional state-of-the-art algorithms is challenging. To address this gap we employ the Globally Optimal Iterative Closest Point (GO-ICP) algorithm to compute correspondences, and a fast approximation algorithm (FastSpectrum) to ex- tract the eigenvectors of the Laplace-Beltrami operator for creating functional maps. Finally, by visualizing the distortion of these maps we identify changes in the coral reefs over time. Our approach is fully automatic, does not require user specified landmarks or an initial map, and surpasses competing shape correspondence methods on coral reef models. Furthermore, our analysis has detected the changes manually marked by humans, as well as additional changes at a smaller scale that were missed during manual inspection. We have additionally used our system to analyse a coral reef model that was too extensive for manual analysis, and validated that the changes identified by the system were correct.\nM.Sc. student under the supervision of Prof. Miri Ben-Chen.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 401&nbsp;
UID:eventx6a5a287eec90f10757
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20250219T113000
DTEND;TZID=Asia/Jerusalem:20250219T123000
DTSTAMP;TZID=Asia/Jerusalem:20250219T113000
SUMMARY: ceClub  talk by Prof. Rafael Pass (Tel-Aviv University)  CE-Club - On Cryptography and Kolmogorov Complexity  at 2025-02-19 11:30:00
DESCRIPTION:Whether secure Cryptography exists is one of the most important open problems in Computer Science: Cryptographic schemes today rely on unproven computational hardness assumption.\nWe will survey a recent thread of work (Liu-Pass,FOCS&rsquo;20, Liu-Pass-STOC'21,.., Ball-Liu-Pass-Mazor, FOCS&rsquo;23, Liu-Pass&rsquo;EUROCRYPTO&rsquo;24) showing *equivalences* between the existence of some of the most basic cryptographic primitives, and the hardness of various computational problems related to the notion of *time-bounded Kolmogorov Complexity* (dating back to the 1960s).\nThese results yield the first natural computational problems *characterizing* the feasibility of central primitives and protocols in Cryptography, as well as the first *unstructured* computational problems enabling public-key cryptography.\nNo prior knowledge of Cryptography or Kolmogorov complexity will be assumed.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Meyer 1061 &amp; Zoom
UID:eventx6a5a287eec9d310754
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20250220T090000
DTEND;TZID=Asia/Jerusalem:20250220T160000
DTSTAMP;TZID=Asia/Jerusalem:20250220T090000
SUMMARY: CSpecial Event  Open Day at the Technion 20.2.25  at 2025-02-20 09:00:00
DESCRIPTION:Interested in undergraduate studies?\nInformation about undergraduate study tracks at the Faculty of Computer Science at the Technion at the link\nAn open day at the Technion will be held on February 20, 2025\nTo register for the open day at the link
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Churchill Building &amp; Taub 1
UID:eventx6a5a287eec9e710749
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20250226T100000
DTEND;TZID=Asia/Jerusalem:20250226T120000
DTSTAMP;TZID=Asia/Jerusalem:20250226T100000
SUMMARY: CSpecial Talk  by Leopold Grinberg (AMD & Brown University)   AMD GPU Roadmap, AI Libraries and Software Optimization  at 2025-02-26 10:00:00
DESCRIPTION:Leopold Grinberg is a Fellow Software Systems Design Engineer at Advanced Micro Devices (AMD) and an Adjunct Lecturer in Applied Mathematics at Brown University.\nHe earned his PhD in Applied Mathematics from Brown University in 2009 and holds a master&rsquo;s degree in Mechanical Engineering from Ben-Gurion University of the Negev, awarded in 2003\nHe focuses on High Performance Computing - systems and applications. Specifically, his current objective is to&nbsp;optimize application performance and develop parallel algorithms for software running on the AMD CPUs and GPUs.\nAgenda\n1- AMD Overview and technology Update &ndash; (Server, Network, Accelerators, FPGA's ) &nbsp;- Yigal Shamayev, Sr Business Development Executive &nbsp;| &nbsp;AMD\n2 - AMD Overview and GPU Roadmap &ndash; &nbsp;Dr&rsquo; Leopold Grinberg\n3 - AMD RocM and AI libraries Software dev and support &ndash; &nbsp;Dr&rsquo; Leopold Grinberg\n4 - Open Questions&nbsp;\n&nbsp;
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Room 337
UID:eventx6a5a287eec9f810756
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20250226T113000
DTEND;TZID=Asia/Jerusalem:20250226T123000
DTSTAMP;TZID=Asia/Jerusalem:20250226T113000
SUMMARY: ceClub  talk by Dr. Daniel Amir (Technion)  CE-Club - Oblivious Reconfigurable Datacenter Networks  at 2025-02-26 11:30:00
DESCRIPTION:As Moore's Law slows down, packet switch capabilities are falling behind datacenter demands. This has made optical circuit switches increasingly attractive in datacenter networks. These switches have already seen significant commercial use in the form of hybrid networks, which combine both packet switches and circuit switches. Recent advances in optical circuit switching technology can now operate fast enough to potentially fully replace packet switches, when combined with novel network designs.\nThis talk presents my research into Oblivious Reconfigurable Networks (ORNs), a design paradigm capable of using the full capabilities of emerging fast circuit switches. I will describe Shale, an ORN that achieves a tunable tradeoff between throughput and latency which is Pareto optimal for ORNs. Along the way, I will also touch on the current state of the art in commercially-deployed hybrid networks. Finally, I will discuss our present research into Semi-Oblivious Reconfigurable Networks (SORNs), which extend ORNs with intuitions found in commercial hybrid networks to further improve the performance possible on fast circuit-switched networks.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Meyer 1061 &amp; Zoom
UID:eventx6a5a287eeca0b10760
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20250302T143000
DTEND;TZID=Asia/Jerusalem:20250302T153000
DTSTAMP;TZID=Asia/Jerusalem:20250302T143000
SUMMARY: MSC  talk by Ohad Einav  A market for accuracy: Classification under Competition  at 2025-03-02 14:30:00
DESCRIPTION:Machine learning models play a key role for service providers looking to gain market share in consumer markets. However, traditional learning approaches do not take into account the existence of additional providers, who compete with each other for consumers. Our work aims to study learning in this market setting, as it affects providers, consumers, and the market itself.\nWe begin by analyzing such markets through the lens of the learning objective, and show that accuracy cannot be the only consideration. We then propose a method for classification under competition, so that a learner can maximize market share in the presence of competitors.\nWe show that our approach benefits the providers as well as the consumers, and find that the timing of market entry and model updates can &nbsp;be crucial. We display the effectiveness of our approach across a range of domains, from simple distributions to noisy datasets, and show that the market as a whole remains stable by converging quickly to &nbsp;an equilibrium.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 401 &amp; Zoom
UID:eventx6a5a287eeca1d10755
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20250304T113000
DTEND;TZID=Asia/Jerusalem:20250304T123000
DTSTAMP;TZID=Asia/Jerusalem:20250304T113000
SUMMARY: PHD  talk by Guy Ohayon  Image Restoration and Compression with Generative Models: Theory and Practice  at 2025-03-04 11:30:00
DESCRIPTION:In this seminar, I will discuss several fundamental challenges and limitations associated with high-perceptual-quality image restoration methods, and propose practical restoration and compression schemes. Specifically, I will first examine deterministic image restoration algorithms and show why striving for high output quality while maintaining consistency with the input measurements inevitably leads to algorithmic instability and vulnerability to adversarial attacks.\nSecondly, since the perceptual quality and distortion of the reconstructions are typically at odds with each other, a key challenge in image restoration is to minimize the distortion under a constraint of perfect output quality. To address this optimization problem, I will introduce a novel algorithm that leverages a rectified flow model to approximate the optimal solution.\nFinally, I will present an innovative generative approach based on pre-trained diffusion models, which produces high-quality image samples along with their losslessly compressed bit-stream representations. This new generative framework seamlessly extends to a variety of tasks, including image compression, compressed image restoration, compressed image editing, and more generally, any compressed conditional generation task.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 6 &amp; Zoom
UID:eventx6a5a287eeca2f10758
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20250309T093000
DTEND;TZID=Asia/Jerusalem:20250309T103000
DTSTAMP;TZID=Asia/Jerusalem:20250309T093000
SUMMARY: PHD  talk by Michal Edelstein  Geometry Processing Techniques for Mesh Realization and Analysis  at 2025-03-09 09:30:00
DESCRIPTION:Mesh realization focuses on developing algorithms for fabricating digital curved shapes in the real world. The goal is to derive a framework for realizing 3D shapes using various materials &mdash; such as wood, paper, or yarn &mdash; while keeping the process as automatic as possible and still allowing the user to make design choices. Material properties (i.e., their possible local deformations) guide the mathematical formulation of the underlying optimization problems.\nWe focus on two central methods. First, we introduce a technique for computing planar hexagonal meshes, an approach particularly useful in architectural contexts where components must be cut from flat materials. Second, we present an automatic algorithm for generating viable crochet instructions directly from 3D models. We describe the mathematical formulations and optimization strategies, including the constraints and objectives, that are used to achieve the desired results.\nBeyond these fabrication-focused approaches, we also develop additional mesh processing techniques needed to support them. These include a general framework for computing regularized geodesic distances and an efficient optimization algorithm capable of handling various regularizers, which can be used to improve shape quality in the realization process. These methods serve as a toolkit for both mesh processing and physical fabrication.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 401
UID:eventx6a5a287eeca4210759
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20250312T120000
DTEND;TZID=Asia/Jerusalem:20250312T130000
DTSTAMP;TZID=Asia/Jerusalem:20250312T120000
SUMMARY: PHD  talk by Yaniv Nemcovsky  Adversarial Attacks In The Real World  at 2025-03-12 12:00:00
DESCRIPTION:Deep neural networks are known to be susceptible to adversarial perturbations, small perturbations that alter the network's output and exist under strict norm limitations. Universal adversarial perturbations aim to alter the model's output on a set of out-of-sample data and present a more realistic use case, as awareness of the model's exact input is not required. In addition, patch adversarial attacks denote the setting where the adversarial pertubations are limited to consist of patches with a given shape and number. This work studies realistic applications of adversarial attacks on visual-based models and the robustness of inference-based defenses. We first consider a randomized smoothing-based defense and show that adversarial attacks can generalize to distributions of inputs and models. We then optimize physical passive patch universal adversarial attacks on visual odometry-based autonomous navigation systems. A patch adversarial perturbation poses a severe security issue for such navigation systems and can mislead them onto some collision course. Finally, we consider the optimal placement of multiple such patches and, to the best of our knowledge, present the first direct solution to optimizing the locations and pertubations of multiple patches.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 601 &amp; Zoom
UID:eventx6a5a287eeca5210761
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20250317T133000
DTEND;TZID=Asia/Jerusalem:20250317T143000
DTSTAMP;TZID=Asia/Jerusalem:20250317T133000
SUMMARY: MSC  talk by Tomer Katz  SCOOPY: Enhancing Program Synthesis with Hierarchical Example Specifications  at 2025-03-17 13:30:00
DESCRIPTION:As program synthesizers become integrated into IDEs, programmers combine synthesized code and manually written&nbsp;code within the same project. We therefore built a Programming-by-Example (PBE) synthesizer that&nbsp;documents the example specifications provided to it alongside the result snippet that satisfies them.\nWe also&nbsp;modified the IDE to treat these example scopes as localized tests for the code they surround, in case they or the&nbsp;code are edited. Unless strict limitations are imposed on how users can edit example scopes, and where they&nbsp;can call the synthesizer, scopes of example-specified code can wind up encompassing others.\nThe programmer can decide to send such a hierarchical scope to the synthesizer, to refactor the code, to automatically correct manually-written code, or simply to fix a mistake found in an example. State of the art PBE synthesizers cannot handle this hierarchical specification: synthesis will only consider the outer-most block, discarding the user intention contained in inner scopes. This can lead to undesired, overfitted programs.\nTo address this information loss we propose Spec Scooping, a syntax-guided technique to &ldquo;scoop&rdquo; out and&nbsp;preserve the intent from inner example scopes. We accompany Spec Scooping with a host of IDE features to help&nbsp;programmers edit example scopes and the code inside them, including an identification of when specifications&nbsp;contradict each other. Since Spec Scooping enriches the specifications sent to the synthesizer, it also requires&nbsp;modifications to the bottom-up Observational Equivalence synthesis algorithm.\nWe implement Spec Scooping in a tool ScooPy, including a development environment that supports scooping&nbsp;and an extended synthesizer.We evaluate ScooPy on 33 benchmarks based on SyGuS competition benchmarks.\nOur results show that hierarchical specifications are generally more expressive, and that, compared to nonhierarchical&nbsp;specifications at the same level of expressiveness, scooping can provide a performance boost. We&nbsp;also performed two small-scale qualitative studies of ScooPy to gauge the benefits of attaching the specifications&nbsp;to a synthesis result and of users&rsquo; interaction with ScooPy&rsquo;s development environment.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 601 &amp; Zoom
UID:eventx6a5a287eeca6410753
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20250318T113000
DTEND;TZID=Asia/Jerusalem:20250318T133000
DTSTAMP;TZID=Asia/Jerusalem:20250318T113000
SUMMARY: pixel-club  talk by Eyal Hanania  Pixel Club - Model-Based Self-Supervised Motion Correction for Robust Cardiac T1 Mapping  at 2025-03-18 11:30:00
DESCRIPTION:Cardiac T1 mapping is a crucial quantitative MRI technique for diagnosing diffuse myocardial diseases. Traditional methods rely on breath-hold sequences and ECG-based cardiac triggering, but these approaches face challenges with patient compliance, limiting their clinical effectiveness. Image registration can enable motion-robust T1 mapping, but intensity variations between time points complicate the process.\nWe introduce MBSS-T1, a subject-specific self-supervised model for motion correction in cardiac T1 mapping. Our method incorporates physical constraints for enforcing signal decay behavior and anatomical constraints ensuring realistic deformations along the longitudinal relaxation axis. We evaluated MBSS-T1 on a public dataset of 210 patients (STONE sequence) and a prospective dataset of 19 subjects acquired at the Technion&rsquo;s May-Blum-Dahl Human MRI research center. The model outperformed baseline deep-learning registration methods in model fitting quality, anatomical alignment, and clinical assessment scores provided by expert radiologists.\nMBSS-T1 enables motion-robust cardiac T1 mapping for both free-breathing and breath-hold acquisitions, improving accessibility for a broader range of patients while reducing the reliance on large annotated datasets.Our work was presented and published at MICCAI 2023 and ISMRM 2024 (oral), as well as in the journal Medical Image Analysis (2025). It was honored with the MICCAI STAR Award (2023) and the ISMRM 2024 Summa Cum Laude Merit Award.\nEyal is currently pursuing an MSc in Electrical and Computer Engineering at the Technion, under the joint supervision of Dr. Moti Freiman (Faculty of Biomedical Engineering) and Prof. Israel Cohen (Faculty of Electrical and Computer Engineering). His research focuses on developing deep-learning models with physical constraints for motion correction in medical imaging.\nIn addition to his academic work, Eyal is a Deep Learning and Computer Vision Algorithm Engineer at WSC Sports. Previously, he worked at GE Research. He earned his BSc in Electrical and Computer Engineering from the Technion.M.Sc. student under the supervision of Prof. Moti Freiman and Prof. Israel Cohen.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom &amp; Meyer 1061
UID:eventx6a5a287eeca7710762
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20250324T113000
DTEND;TZID=Asia/Jerusalem:20250324T123000
DTSTAMP;TZID=Asia/Jerusalem:20250324T113000
SUMMARY: MSC  talk by Idan Kashani  Direct Representation of Large Language Models in the Semantic Task Space  at 2025-03-24 11:30:00
DESCRIPTION:The open-source community offers a vast and continually expanding array of large language models (LLMs), accompanied by diverse benchmarks to evaluate their performance. While this wealth ecosystem provides users with many models that may align with their objectives, the sheer number of options makes selection complex and time-consuming. A model may appear proficient in a given domain, yet underperform on specific instances.\nWe introduce a straightforward, efficient, and scalable linear method for creating structured representations of models by leveraging the diversity of existing benchmarks. Our approach is highly interpretable and requires no training. This makes it particularly well-suited for dynamic environments where models and datasets are frequently updated. By embedding models into a task-oriented space, our method facilitates systematic retrieval based on predefined properties, such as performance. We apply this method within a library of models and datasets, storing the representations as model metadata, and demonstrate their practical utility for success prediction and model selection.\nAdditionally, we present Hysteresis Rectified Linear Unit (HeLU), a novel activation function, designed to address the "dying ReLU" problem. During inference, HeLU functions identically to the Rectified Linear Unit (ReLU), preserving computational efficiency. We show that HeLU provides a lightweight alternative to more complex activation functions, offering a practical trade-off between performance and computational cost.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 601 &amp; Zoom&nbsp;
UID:eventx6a5a287eeca8b10764
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20250325T113000
DTEND;TZID=Asia/Jerusalem:20250325T133000
DTSTAMP;TZID=Asia/Jerusalem:20250325T113000
SUMMARY: pixel-club  talk by Sagi Monin  Pixel Club - Shaping Light From Microscopy To Holographic Displays  at 2025-03-25 11:30:00
DESCRIPTION:Shaping light plays a crucial role in science and technology, enabled by advancements in spatial-light modulator (SLM) technology. In this talk, we will explore two key applications: WFS microscopy for deep tissue imaging and holographic displays.\nIn the first part, we will discuss wavefront shaping systems and how they enable deep tissue imaging by correcting aberrations caused by tissue inhomogeneity.&nbsp;However, estimating optimal modulations remains challenging due to unknown tissue structures. Most current techniques employ slow coordinate descent algorithms, which sequentially scan all modulation elements and query their values independently. Thus, their complexity scales prohibitively with the number of modulation parameters. We present a rapid wavefront shaping method that transitions from coordinate to gradient descent optimization, simultaneously updating all modulation parameters. Our approach employs a non-invasive, guide-star-free score function to quantify modulation effectiveness.\nWe derive an analytical framework expressing the score&rsquo;s gradient with respect to all parameters. Although this gradient depends on the unknown tissue structure, we demonstrate how it can be inferred from optical measurements. This method enables rapid computation of high-resolution wavefront corrections required for thick tissue samples, with complexity independent of the number of modulation parameters. Finally, we demonstrate our framework&rsquo;s efficacy in correcting aberrations in a coherent confocal microscope.\nIn the second part, we explore holographic displays and their limitation: etendue. Etendue is defined as the product of the display&rsquo;s size and angular range, which is bounded by pixel count. Current SLMs are inadequate for realistic displays, needing far more pixels. Existing strategies for etendue expansion use a diffractive optical element (DOE), whose pitch is smaller than that of the SLM, thereby spreading light over a wider angle. However, a fixed phase mask does not increase degrees of freedom, resulting in loss of image quality. We study the trade-offs of static phase masks for etendue expansion. We characterize what trade-offs are involved, and how specific phase mask design could support better holograms.&nbsp;\nNext, we suggest expanding the etendue by augmenting the pixel with tilting capabilities. We show that tiltable displays can be realized using a cascade of binary tilt layers, each capable of tilting the light towards one of two orientations. We implement a proof-of-concept display and demonstrate its applicability for displaying multi-view or holographic content with increased size and angular field-of-view.\nSagi Monin is a Ph.D. student under the supervision of Prof. Anat Levin . His research focuses on computational photography, wavefront-shaping for microscopy. and holographic displays.He received the Jacobs-Qualcomm scholarship, the Sherman scholarship, the Boaz Porat award, and Meyer excellence award.&nbsp;PhD student under the supervisor of Prof. Anat Levin
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:506, Zisapel Building &amp; Zoom
UID:eventx6a5a287eeca9d10767
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20250326T103000
DTEND;TZID=Asia/Jerusalem:20250326T113000
DTSTAMP;TZID=Asia/Jerusalem:20250326T103000
SUMMARY: PHD  talk by Amit Bracha   Correspondence, Classification and Reconstruction of shapes in 3D  at 2025-03-26 10:30:00
DESCRIPTION:Surface reconstruction and shape matching are critical challenges in 3D geometry, underpinning the creation of detailed digital models and enabling the robust alignment of complex shapes - capabilities that drive advancements in computer vision, robotics, and medical imaging.\nIn this talk, we introduce a method that leverages a novel view synthesis algorithm (3DGS) to reconstruct 3D surfaces from real-world data, outperforming existing techniques. We then tackle the challenge of analyzing such reconstructions, where a common issue is that only part of an object is visible - which complicates shape analysis and in particular shape correspondence.\nTraditional methods often fail to produce reliable matches under these conditions. Through both theoretical and quantitative examinations of prior approaches, we identified critical flaws in their workflows, which motivated the development of a new pipeline specifically for partial shape correspondence.\nFurthermore, to mitigate errors throughout the learning process, we propose a new loss function based on the Gromov-Wasserstein distance that by utilizing both geodesic and Euclidean distances effectively excludes geometric distortions introduced by missing parts of the shape.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 401 &amp; Zoom
UID:eventx6a5a287eecab510765
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20250331T123000
DTEND;TZID=Asia/Jerusalem:20250331T133000
DTSTAMP;TZID=Asia/Jerusalem:20250331T123000
SUMMARY: MSC  talk by Dvir David Biton  Geometric Sketch: The Inflatable-Shrinkable Sketch  at 2025-03-31 12:30:00
DESCRIPTION:Stream frequency measurements are fundamental in many data stream applications such as financial data trackers, intrusion detection systems, and network monitoring. Count-Min Sketch and its variants have been widely adopted for this task due to their space efficiency and ability to provide approximate frequency estimates with bounded error guarantees. However, these sketches suffer from a limitation: they have a fixed memory usage determined by the desired accuracy, unable to adapt to changing memory availability or accuracy requirements during runtime. Dynamic solutions exist, but they are likewise still constrained. We introduce the Geometric Sketch, a frequency estimation sketch that addresses the limitations of existing sketches by offering dynamic memory allocation. Unlike existing sketches with fixed memory footprints, the Geometric Sketch can dynamically grow or shrink its memory usage at an arbitrary size, down to a single-counter granularity, making it well-suited to systems with changing memory availability When evaluated using real Internet packet traces, the Geometric Sketch achieves higher accuracy compared to the existing method, DCMS, by employing fine-grained expansion and compression.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 301 &amp; Zoom
UID:eventx6a5a287eecac710768
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20250401T110000
DTEND;TZID=Asia/Jerusalem:20250401T120000
DTSTAMP;TZID=Asia/Jerusalem:20250401T110000
SUMMARY: MSC  talk by Nadav Sellam  Experiment-Guided Generative Models For Protein Structure & Dynamics  at 2025-04-01 11:00:00
DESCRIPTION:Proteins exist as a dynamic ensemble of multiple conformations, and these motions are often crucial for their functions. However, current structure prediction methods predominantly yield a single conformation, overlooking the conformational heterogeneity revealed by diverse experimental modalities.\nThis work presents a framework for building experiment-grounded protein structure generative models that infer conformational ensembles consistent with measured experimental data. The key idea is to treat state-of-the-art protein structure predictors (e.g., AlphaFold3) as sequence-conditioned structural priors, and cast ensemble modeling as posterior inference of protein structures given experimental measurements.\nThrough extensive real-data experiments, we demonstrate the generality of our method to incorporate a variety of experimental measurements. In particular, our framework uncovers previously unmodeled conformational heterogeneity from crystallographic densities, and generates high-accuracy NMR ensembles orders of magnitude faster than the status quo. Notably, we demonstrate that our ensembles outperform AlphaFold3 (Abramson et al., 2024) and sometimes better fit experimental data than publicly deposited structures to the Protein Data Bank (PDB, Burley et al. (2017)).\nWe believe that this approach will unlock building predictive models that fully embrace experimentally observed conformational diversity.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom
UID:eventx6a5a287eecad810770
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20250401T113000
DTEND;TZID=Asia/Jerusalem:20250401T133000
DTSTAMP;TZID=Asia/Jerusalem:20250401T113000
SUMMARY: pixel-club  talk by Eitan Richardson (Lightricks, The Hebrew University)  Pixel Club - LTX-Video: Realtime Video Latent Diffusion  at 2025-04-01 11:30:00
DESCRIPTION:We introduce LTX-Video, a transformer-based latent diffusion model that adopts a holistic approach to video generation by seamlessly integrating the responsibilities of the Video-VAE and the denoising transformer.\nUnlike existing methods, which treat these components as independent, LTX-Video aims to optimize their interaction for improved efficiency and quality. At its core is a carefully designed Video-VAE that achieves a high compression ratio of 1:192, with spatiotemporal downscaling of 32 x 32 x 8 pixels per token, enabled by relocating the patchifying operation from the transformer&rsquo;s input to the VAE&rsquo;s input.\nOperating in this highly compressed latent space enables the transformer to efficiently perform full spatiotemporal self-attention, which is essential for generating high-resolution videos with temporal consistency.\nHowever, the high compression inherently limits the representation of fine details. To address this, our VAE decoder is tasked with both latent-to-pixel conversion and the final denoising step, producing the clean result directly in pixel space. This approach preserves the ability to generate fine details without incurring the runtime cost of a separate upsampling module.\nOur model supports diverse use cases, including text-to-video and image-to-video generation, with both capabilities trained simultaneously. It achieves faster-than-real-time generation, producing 5 seconds of 24 fps video at 768&times;512 resolution in just 2 seconds on an Nvidia H100 GPU, outperforming all existing models of similar scale.\nThe source code and pre-trained models are publicly available, setting a new benchmark for accessible and scalable video generation.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:506 Zisapel Building &amp; Zoom
UID:eventx6a5a287eecae910771
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20250402T113000
DTEND;TZID=Asia/Jerusalem:20250402T123000
DTSTAMP;TZID=Asia/Jerusalem:20250402T113000
SUMMARY: ceClub  talk by Dr. Nadav Amit (Technion)  CE-Club - When a File Means a File: Proper Huge Pages for Code  at 2025-04-02 11:30:00
DESCRIPTION:Despite huge pages dramatically reducing CPU frontend stalls from address translation, their use for executable code remains limited due to operating system constraints and impracticality of rebuilding system binaries with special alignment. Current solutions that copy code into huge pages break essential system functionality - preventing memory sharing between processes, disrupting debugging tools, and interfering with memory management operations.\nIn this talk, I will present a practical userspace solution that achieves huge page performance benefits while preserving critical system services. Our approach transforms binaries to align code segments with huge page boundaries post-linkage while maintaining all internal references, and orchestrates page cache operations to ensure proper mapping. PostgreSQL evaluations demonstrate up to 7% performance improvement through a 94% reduction in iTLB misses, while maintaining memory sharing, debugging support, and proper memory management
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Meyer building 1061 &amp; Zoom
UID:eventx6a5a287eecafc10775
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20250402T153000
DTEND;TZID=Asia/Jerusalem:20250402T163000
DTSTAMP;TZID=Asia/Jerusalem:20250402T153000
SUMMARY: MSC  talk by Rotem Shavitt  Finding Possible Winners in Spatial Voting with Incomplete Information  at 2025-04-02 15:30:00
DESCRIPTION:We consider a spatial voting model where both candidates and voters are positioned in the d-dimensional Euclidean space, and each voter ranks candidates based on their proximity to the voter's ideal point. We focus on the scenario where the given information about the locations of the voters' ideal points is incomplete; for each dimension, only an interval of possible values is known.\nIn this context, we investigate the computational complexity of determining the possible winners under positional scoring rules. Our results show that the possible winner problem in one dimension is solvable in polynomial time for all k-truncated voting rules with constant k. Moreover, for some scoring rules for which the possible winner problem is NP-complete, such as approval voting for any dimension or k-approval for d &gt;= 2 dimensions, we give an FPT algorithm parameterized by the number of candidates. Finally, we classify tractable and intractable settings of the weighted possible winner problem in one dimension, and resolve the computational complexity of the weighted case for all two-valued positional scoring rules when d=1.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 401 &amp; Zoom&nbsp;
UID:eventx6a5a287eecb0b10763
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20250406T143000
DTEND;TZID=Asia/Jerusalem:20250406T153000
DTSTAMP;TZID=Asia/Jerusalem:20250406T143000
SUMMARY: MSC  talk by Omer Yerushalmi  The Capacity of the Weighted Read Channel  at 2025-04-06 14:30:00
DESCRIPTION:Nanopore sequencing is emerging as a powerful tool for storing digital information in DNA molecules. This technique offers several advantages over traditional methods, making it an attractive area of research. In this work, we focus on a simplified model of&nbsp;the nanopore sequencing process, represented as a channel.\nThis channel operates by taking a DNA sequence and analyzing it one segment at a time. It uses a sliding window of a specific length, denoted by ℓ, to scan the sequence. The window is then shifted by&nbsp;&delta; characters, and the process repeats. The output of the channel called the read vector, is a collection of values where each value represents the sum of the bases (like A, C, G, T) within a particular window.\nThe channel&rsquo;s capacity signifies the maximum rate of information transmission through it. Prior research has established capacity values for specific combinations of ℓ and &delta;. In this study, we delve deeper, demonstrating that when &delta; &lt; ℓ &lt; 2&delta;, the channel&rsquo;s capacity can be expressed as (1/&delta;) log (1/2 (ℓ + 1 +&radic;((ℓ + 1)2 &minus; 4(ℓ &minus; &delta;)(ℓ &minus; &delta; + 1))). Furthermore, we establish an upper bound on the&nbsp;capacity when 2&delta; is less than ℓ. Finally, to enhance the model&rsquo;s complexity, we extendit to a two-dimensional scenario and present various findings on its capacity. This extended model brings us closer to mimicking the real intricacies of nanopore sequencing and its potential for DNA storage
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 601 &amp; Zoom&nbsp;
UID:eventx6a5a287eecb1d10774
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20250408T110000
DTEND;TZID=Asia/Jerusalem:20250408T120000
DTSTAMP;TZID=Asia/Jerusalem:20250408T110000
SUMMARY: MSC  talk by Ron Kantorovich  Drug Target Prediction by Learning from High-Throughput Metabolomics Data and Metabolic Pathway Structures  at 2025-04-08 11:00:00
DESCRIPTION:OMICS-based screening offers a promising approach for untargeted drug discovery. Mass-spectrometry metabolomics and proteomics were recently used for inferring drug mechanisms of action and off-target effects, analyzing the cellular response to treatment with numerous clinically approved drugs and tool compounds. In this talk, we present a novel high-throughput LC-MS metabolomics screening pipeline and a deep-learning method specifically designed for cellular metabolism.\nOur novel Graph Neural Network (GNN) model fits the unique structure of this domain, enabling accurate identification of target pathways. Learning from time- and dose-dependent metabolic responses of cultured cells treated with 76 known metabolic inhibitors, we correctly identified the target pathway within the top three ranked pathways for ~50% of the drugs. Applying this approach to a diversity library of 1,020 drug-like compounds, we discovered five novel inhibitors targeting clinically relevant pathways and enzymes involved in purine and pyrimidine biosynthesis and redox metabolism. Our pipeline is readily scalable for screening thousands of compounds to identify new, clinically relevant metabolic inhibitors.\nWe will dive deep into our machine learning approach, offering valuable insights and comprehensive ablation studies that highlight its strength and domain-specific innovation.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 401 &amp; Zoom
UID:eventx6a5a287eecb2d10769
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20250408T113000
DTEND;TZID=Asia/Jerusalem:20250408T133000
DTSTAMP;TZID=Asia/Jerusalem:20250408T113000
SUMMARY: pixel-club  talk by Hila Chefer (Tel Aviv University)  Pixel Club - Toward Generative Models that Understand the Visual World  at 2025-04-08 11:30:00
DESCRIPTION:Despite remarkable advances, visual generative models are still far from faithfully modeling the world, struggling with fundamental aspects such as spatial relations, physics, motion, and dynamic interactions.\nIn this talk, I present a line of work that tackles these challenges, based on a deep understanding of the inner mechanisms that drive models. I will begin by analyzing state-of-the-art visual generators, gaining insights into the underlying reasons for their limited understanding. Building upon these insights, I will demonstrate methods that significantly enhance both spatial and temporal reasoning in image and video generation, surpassing even resource-intensive proprietary models without relying on additional data or model scaling. I will conclude the talk by discussing open challenges and future directions for advancing faithful world modeling in visual generative models.\nBio:Hila is a PhD candidate at Tel Aviv University, advised by Prof. Lior Wolf. Her research focuses on understanding, interpreting, and correcting the predictions of deep foundational models. During her PhD, she was a visiting researcher at Google Research, Google DeepMind, and Meta AI, where she led works on video generation.\nHila has received several awards, including the Fulbright Postdoctoral Fellowship, the Eric and Wendy Schmidt Postdoctoral Award, the Deutsch Prize for Outstanding PhD Students, and the Council for Higher Education (VATAT) Award for Outstanding PhD Students.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:506, Zisapel Building &amp; Zoom&nbsp;
UID:eventx6a5a287eecb3d10777
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20250409T113000
DTEND;TZID=Asia/Jerusalem:20250409T123000
DTSTAMP;TZID=Asia/Jerusalem:20250409T113000
SUMMARY: PHD  talk by Sanketh Vedula  On Geometric Learning, Statistical Inference, and Biomolecular Modeling  at 2025-04-09 11:30:00
DESCRIPTION:In this talk, I present three research directions from my PhD.\n1. Geometric Learning for Structured Data. Firstly, we introduce a simple, spectral-geometric approach for matrix completion on graphs. Our approach couples the priors implicitly induced by gradient descent with explicitly imposed spectral-geometric priors and achieves strong performance in drug-target interaction prediction and recommendation systems applications. Secondly, we introduce inductive/generalizable solvers for quadratic optimal transport problems, demonstrating superior scalability over transductive methods, with applications in single-cell multi-omic alignment. Finally, on the practical front, we show algorithms for efficient execution of graph machine learning workloads for large-scale recommendation system inference.\n2. Statistical Inference via Optimal Transport. In this line of work, we develop new solvers and extensions for vector quantile regression (VQR), an optimal-transport&ndash;based statistical framework introduced by Carlier et al. 2016. Firstly, we introduce nonlinear VQR, the first practical approach to model quantile functions of multivariate conditional distributions, and demonstrate applications in creating calibrated high-dimensional confidence sets. Secondly, we propose manifold VQR, and extend the notion of conditional quantile functions for manifold-valued response variables. Finally, we introduce continuous solvers for VQR by solving conditional continuous OT problems. We perform fundamental statistical inference tasks on conditional distributions, i.e., sampling, computing likelihoods, constructing confidence sets, computing order statistics (ranks, quantiles, etc.) from the derived OT maps.\n3. Modeling Biomolecules. In this research direction, we make advances in overcoming the "single-sequence, single-structure" dogma in structural biology, and highlight fundamental limitations of protein structure predictors such as AlphaFold. Firstly, we investigate protein structures that exhibit dual conformations in a single crystal, called "altlocs", and identify "stable altlocs", the altloc dual conformations that are thermodynamically stable. We demonstrate that the state-of-the-art protein structure predictors and protein structure generative models fail to recover these dual conformations. We introduce a guided diffusion framework to improve the modeling of biomolecules conditioned on experimental measurements. Secondly, we demonstrate limitations of protein structure prediction algorithms for modeling chimeric proteins. To overcome this, we introduce windowed multiple sequence alignment (MSA), for enriching the MSA of the chimeric proteins to yield improved predictions.&nbsp;
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom
UID:eventx6a5a287eecb4f10778
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20250410T123000
DTEND;TZID=Asia/Jerusalem:20250410T133000
DTSTAMP;TZID=Asia/Jerusalem:20250410T123000
SUMMARY: PHD  talk by Amit Ganz  Randomized and Continuous Algorithms for Submodular Maximization  at 2025-04-10 12:30:00
DESCRIPTION:Submodular functions arise in various fields, including combinatorics, graph theory, information theory, and economics. This thesis addresses two key problems in submodular maximization and presents new algorithmic contributions. The first focuses on the Online Submodular Welfare problem, where bidders with submodular utility functions compete for items arriving sequentially. We propose a randomized algorithm that achieves a tight competitive ratio of 1/4 under adversarial arrivals and improve this to approximately 0.27493 under random arrivals. The second result introduces a Greedy Poisson Process for submodular maximization under matroid constraints, achieving a tight (1 &minus; 1/e &minus; ϵ)-approximation without requiring discretization or rounding. Our algorithm introduces a new framework for submodular maximization under matroid constraints, combining the strengths of both continuous and discrete approaches.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom
UID:eventx6a5a287eecb6210773
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20250410T130000
DTEND;TZID=Asia/Jerusalem:20250410T140000
DTSTAMP;TZID=Asia/Jerusalem:20250410T130000
SUMMARY: MSC  talk by Ron Gatenio  Persistency Race Detection  at 2025-04-10 13:00:00
DESCRIPTION:Nonvolatile Memory (NVM) technologies offer new avenues for building high-performance and crash-consistent applications by combining byte-addressable DRAM-like characteristics with non-volatility. However, these features introduce complex challenges in ensuring data consistency, especially under concurrent access scenarios.\nThis paper introduces PRD, a specialized tool designed to detect persistency races - specific type of concurrency bugs in PM environments that can lead to critical inconsistencies following system failures.\nPRD utilizes graph-theoretical analysis along with happens-before and program-dependence analysis to map causal relationships and dependencies among program operations. While these analyses are well-established techniques in data-race detection, PRD efficiently generalizes across multiple thread interleavings to detect potential races within a single program execution, significantly enhancing the speed and efficiency of race detection in NVM environments.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Room 601 &amp; Zoom
UID:eventx6a5a287eecb7210776
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20250421T123000
DTEND;TZID=Asia/Jerusalem:20250421T143000
DTSTAMP;TZID=Asia/Jerusalem:20250421T123000
SUMMARY: CSpecial Event  Master's Degree Exposure Meeting At The Faculty  at 2025-04-21 12:30:00
DESCRIPTION:You are invited to a master's degree exposure meeting intended for outstanding undergraduates.\nThe meeting will take place on Monday, April 21, 2025 at 12:30 in Auditorium 012 (Floor 0) at the Faculty of Computer Science.\nPlease register at the link by April 15, 2025
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Auditorium 012 (Floor 0)
UID:eventx6a5a287eecb8510772
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20250422T113000
DTEND;TZID=Asia/Jerusalem:20250422T123000
DTSTAMP;TZID=Asia/Jerusalem:20250422T113000
SUMMARY: MSC  talk by Ittay Alfassi  Analysis and Classification for Attacks on Quantum Key Distribution  at 2025-04-22 11:30:00
DESCRIPTION:At the core of practical implementations of Quantum Communication in general and in Quantum Key Distribution (QKD) in particular is the usage of photons and quantum optical devices rather than ideal qubits and qubit gates.The realistic photonic protocol, often combined with non-ideal devices, is vulnerable to various implementation attacks even when the underlying (ideal) protocol is proven secure.\nOur work sheds new light on the said implementation attacks using several new tools which take inspiration from classical cybersecurity by considering the quantum equivalents of vulnerabilities, attack surfaces, and exploits.We define &ldquo;Reversed-Space Attacks&rdquo;, which act as a generic attack surface computation and exploit method against imperfect receivers.\nWe give a concrete quantum-mechanical definition of &ldquo;Quantum Side-Channel Attacks&rdquo;, providing a meaningful distinction from other attack forms.We also define a notion of &ldquo;Quantum Fuzzing&rdquo; as a tool for practical black-box vulnerability research, acting as a complementary tool to Reversed-Space Attacks.\nThe three tools we define are used to analyze multiple known QKD attacks, giving a better understanding of common attack building blocks and relations between attacks.A direct conclusion is that the &ldquo;Bright Illumination&rdquo; attack,which was executed in practice before theoretical predictions, could have been found using our tools and perspective.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 601
UID:eventx6a5a287eecb9410782
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20250422T140000
DTEND;TZID=Asia/Jerusalem:20250422T150000
DTSTAMP;TZID=Asia/Jerusalem:20250422T140000
SUMMARY: MSC  talk by Yarden Lev  Computational Approach to Find Candidates for Cancer Drugs Using Multi-Omics Data  at 2025-04-22 14:00:00
DESCRIPTION:Identifying novel cancer drug candidates requires a comprehensive understanding of drug-induced molecular changes. This study integrates metabolomics and proteomics data with computational methods to investigate the mechanisms of action (MOA) of previously uncharacterized compounds from a large-scale, high-throughput screening experiment. Our approach systematically compares internally generated proteomic profiles with an external dataset of 875 known small molecules, employing computational strategies to mitigate biases inherent in cross-dataset comparisons. Through this multi-omics approach, this study identifies promising cancer drug candidates by uncovering specific molecular targets, potential anticancer mechanisms, and similarities to known compounds with established anticancer effects. By integrating experimental data with computational analyses, this study not only advances the identification of potential cancer therapeutics but also provides a framework for integrating multi-omics data across diverse datasets.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Room 4-17, Emerson Building, Biology &amp; Zoom
UID:eventx6a5a287eecba710781
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20250423T113000
DTEND;TZID=Asia/Jerusalem:20250423T123000
DTSTAMP;TZID=Asia/Jerusalem:20250423T113000
SUMMARY: ceClub  talk by Michael Mirkin (Technion)  CE-Club - Blockchain: Unmotivating, Overpowered, Underperforming — and Still Full of Potential  at 2025-04-23 11:30:00
DESCRIPTION:Blockchain networks like Bitcoin and Ethereum underpin billions in value and promise decentralized trust, yet they face critical challenges: their security depends on fragile economic incentives, their energy consumption raises sustainability concerns, and their limited computational capacity constrains scalability.\nIn this talk, I&rsquo;ll share our research on all three fronts. First, I&rsquo;ll explore incentive vulnerabilities in proof of work that can disrupt participation and threaten system integrity. Then I&rsquo;ll question the &ldquo;more security means more power&rdquo; mantra, showing how shifting costs toward hardware acquisition can achieve the same level of security with far less energy. Finally, I&rsquo;ll present our new approach to extending on chain computation, unlocking fresh possibilities for smarter, more scalable contracts.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Mayer 1061
UID:eventx6a5a287eecbb910784
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20250423T125800
DTEND;TZID=Asia/Jerusalem:20250423T143000
DTSTAMP;TZID=Asia/Jerusalem:20250423T125800
SUMMARY: CSpecial Event  talk by The Piano Auditorium  Dell Technologies Spotlight Day  at 2025-04-23 12:58:00
DESCRIPTION:You are invited to open the semester with DELL for a spotlight day and technology lecture -&nbsp;How Code Builds Your Infrastructure: An Intro to IaCSpeaker, Eli Rosens, Senior Engineer, Product Management, EDGE at Dell Technologies\nWednesday, April 23rd at noon starting at 12:30 in the Piano Auditorium.\nThe event is intended for outstanding students with one year of remaining studies (for graduates of the Spring 2026 semester)\nRegister&nbsp;here\nSee you!
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:
UID:eventx6a5a287eecbc910783
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20250423T130000
DTEND;TZID=Asia/Jerusalem:20250423T140000
DTSTAMP;TZID=Asia/Jerusalem:20250423T130000
SUMMARY: Theory Semina  talk by Rafael Pass (Technion, Cornell Tech, & Cornell)  Theory Seminar: On Cryptography and Kolmogorov Complexity  at 2025-04-23 13:00:00
DESCRIPTION:Whether secure Cryptography exists is one of the most important open problems in Computer Science: Cryptographic schemes today rely on unproven computational hardness assumption.\nWe will survey a recent thread of work (Liu-Pass FOCS&rsquo;20, Liu-Pass STOC&rsquo;21,.., Ball-Liu-Pass-Mazor FOCS&rsquo;23, Liu-Pass EUROCRYPT&rsquo;24) showing *equivalences* between the existence of some of the most basic cryptographic primitives, and the hardness of various computational problems related to the notion of *time-bounded Kolmogorov Complexity* (dating back to the 1960s).\nThese results yield the first natural computational problems *characterizing* the feasibility of central primitives and protocols in Cryptography, as well as the first *unstructured* computational problems enabling public-key cryptography.No prior knowledge of Cryptography or Kolmogorov complexity will be assumed.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 201
UID:eventx6a5a287eecbda10787
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20250424T123000
DTEND;TZID=Asia/Jerusalem:20250424T133000
DTSTAMP;TZID=Asia/Jerusalem:20250424T123000
SUMMARY: MSC  talk by Benjamin A. Cohen  Benchmarking Ophthalmology Foundation Models for Clinically Significant AMD Detection  at 2025-04-24 12:30:00
DESCRIPTION:Self-supervised learning (SSL) has enabled Vision Transformers (ViTs) to learn robust representations from large-scale natural image datasets, enhancing their generalization across domains. In retinal imaging, foundation models pretrained on either natural or ophthalmic data have shown promise, but the benefits of in-domain pretraining remain uncertain. To investigate this, we benchmark six SSL-pretrained ViTs on seven digital fundus image (DFI) datasets totaling 70,000 expert-annotated images for the task of moderate-to-late age-related macular degeneration (AMD) identification. Our results show that iBOT pretrained on natural images, achieves the highest out-of-distribution generalization, with AUROCs of 0.80&ndash;0.97, outperforming domain-specific models, which achieved AUROCs of 0.78&ndash;0.96 and a baseline ViT-L with no pretraining, which achieved AUROC of 0.68-0.91. These findings highlight the value of foundation models in improving AMD identification, and challenge the assumption that in-domain pretraining is necessary. Furthermore, we release BRAMD, an open-access dataset (n=587) of DFIs with AMD labels from Brazil.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Bio-Medical Engineering Room 201 &amp; Zoom&nbsp;
UID:eventx6a5a287eecbea10786
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20250427T170000
DTEND;TZID=Asia/Jerusalem:20250427T180000
DTSTAMP;TZID=Asia/Jerusalem:20250427T170000
SUMMARY: MSC  talk by Avidan Borisov  Batching with End-to-End Performance Estimation  at 2025-04-27 17:00:00
DESCRIPTION:Batching heuristics are used in multiple layers of the TCP/IP stack, attempting to improve performance by amortizing overheads. When defining "performance" as average latency and throughput, optimal batching decisions can be infeasible if application-perceived end-to-end performance is unknown, which is commonly the case in general-purpose setups. We propose to address this problem by occasionally adding several easily-maintained counters to TCP metadata exchanges and using them to estimate end-to-end performance via Little's law.\nWe contend that this approach can yield reasonable estimates without application support, and that applications can optionally further improve accuracy through a trivial interface we introduce. We demonstrate experimentally that such estimates have the potential to significantly improve the quality of batching decisions, extending Redis's range of sustainable throughput with tolerable latencies by more than 1.6x and improving the latencies within this range by up to 3.9x.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Room 601 &amp; Zoom
UID:eventx6a5a287eecbfd10785
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20250428T113000
DTEND;TZID=Asia/Jerusalem:20250428T123000
DTSTAMP;TZID=Asia/Jerusalem:20250428T113000
SUMMARY: MSC  talk by Ron Benchetrit  Anytime incremental rhoPOMDP planning in continuous spaces  at 2025-04-28 11:30:00
DESCRIPTION:Partially Observable Markov Decision Processes (POMDPs) provide a robust framework for decision-making under uncertainty in applications such as autonomous driving and robotic exploration. Their extension, rhoPOMDPs, introduces belief-dependent rewards, enabling explicit reasoning about uncertainty. Existing online rhoPOMDP solvers for continuous spaces rely on fixed belief representations, limiting adaptability and refinement - critical for tasks such as information-gathering. We present rhoPOMCPOW, an anytime solver that dynamically refines belief representations, with formal guarantees of improvement over time. To mitigate the high computational cost of updating belief-dependent rewards, we propose a novel incremental computation approach. We demonstrate its effectiveness for common entropy estimators, reducing computational cost by orders of magnitude. Experimental results show that rhoPOMCPOW outperforms state-of-the-art solvers in both efficiency and solution quality.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom
UID:eventx6a5a287eecc0d10788
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20250430T080000
DTEND;TZID=Asia/Jerusalem:20250430T093000
DTSTAMP;TZID=Asia/Jerusalem:20250430T080000
SUMMARY: CSpecial Event  Registration For The 2025 CS Hackathon - Doing Good Is Now Open!!!  at 2025-04-30 08:00:00
DESCRIPTION:The Faculty of Computer Science and the Student Council invite you to do good and be part of the biggest and most significant event of the year!This year's CS Hackathon Doing Good will focus on the topic of "Truth in the Digital Space" - dealing with fake news and disinformation and will take place on Thursday-Friday, May 8-9 at Taub.140 students embarked on a programming marathon with the aim of developing innovative solutions that will help deal with these challenges and contribute to our social and democratic resilience30 hours of teamwork, gaining experience, mentoring from first-class mentors, tons of treats and of course financial prizes for the winners.You are invited to take part and influence the most significant challenge in recent years that threatens public trust and social cohesion. This challenge is already being felt in reality, as you can see here.\nMandatory meetings:April 27 at 5:30 PM &ndash; Pre-Hackathon meeting at the Piano Auditorium in Taub.&bull; Preparatory meeting: Presentation of the rules of the game and Q&amp;A&bull; Lecture: Introduction to Disinformation &ndash; hosted by Idan Ring and Nitzan Yasur, Israeli Internet Association\nHackathon registration &ndash; now open!https://www.cshack-technion.com/registrationRegistration in groups of 3-5 participants | Registration must be made as a group and not as individuals | The number of places is limited. Participation confirmations will be sent by 4/22/2025\nNot sure how to get started? We've got you covered:Looking for an idea? You can get ideas from the list of challenges on the website:https://www.cshack-technion.com/challenges\nDon't have experience? Let's accumulate it together - especially for you we have created short technology workshops that will give you practical tools for Web, AI and NLP development.https://www.cshack-technion.com/courses\nLooking for partners/a group to join? You are welcome to introduce yourself and get to know the dedicated group:https://chat.whatsapp.com/BQbSr5jX1fhDo6Zj1mzVJ5\nQuestions? Talk to us! cshackathon@cs.technion.ac.il\nGood luck to everyone,Cs hack doing good team
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:
UID:eventx6a5a287eecc2010780
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20250504T143000
DTEND;TZID=Asia/Jerusalem:20250504T153000
DTSTAMP;TZID=Asia/Jerusalem:20250504T143000
SUMMARY: MSC  talk by Roee Gross  Synchronization and Indel Correction in Linear Codes  at 2025-05-04 14:30:00
DESCRIPTION:Linear error‑correcting codes that handle insertions and deletions (indels) are fundamental to reliable storage and communication in settings where symbol positions may drift&mdash;from DNA data storage to packet‑based networks. Unlike substitution and erasure errors, indels disrupt sequence alignment and therefore require additional synchronization information.\nThis talk surveys recent constructive techniques that embed such synchronization within fully linear codes, preserving algebraic structure while supporting efficient encoding and decoding. We explain how short synchronization sequences can be interleaved with algebraic‑geometry codes to form &ldquo;half‑linear&rdquo; indel codes, and how a padding&ndash;flattening procedure converts them into fully linear codes over the base field with only a modest rate penalty. A decoder that combines longest‑common‑subsequence matching with asymmetric Hamming decoding then corrects a prescribed fraction of indels in polynomial time.\nAlong the way we discuss a subfield‑Singleton&ndash;style limitation that shows why rate 1/2 cannot be surpassed&mdash;even for a single indel&mdash;when only base‑field linearity is assumed. The resulting framework narrows the gap between known lower and upper bounds for linear indel codes and suggests new directions&mdash;including tighter synchronization primitives and improved high‑rate constructions&mdash;for bringing practical, algebraically structured indel correction closer to theoretical limits.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 601 &amp; Zoom
UID:eventx6a5a287eecc4210790
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20250506T113000
DTEND;TZID=Asia/Jerusalem:20250506T133000
DTSTAMP;TZID=Asia/Jerusalem:20250506T113000
SUMMARY: pixel-club  talk by Chen Katz  Pixel Club - Rolling Waves: A Floating Multi View Network  at 2025-05-06 11:30:00
DESCRIPTION:Environmental sensing has a major gap: three-dimensional (3D) mapping of the atmosphere over the ocean, under clouds. This gap is caused by occlusion of the domain from satellite views and lack of wide-field 3D sensing from space. Moreover, although water covers most of Earth, almost all sensors are currently on dry land. We address the challenge by deriving a new approach: multi-view imaging using a floating network of cameras that have overlapping view fields. Such a setup introduces new computer vision challenges, which we pose and address. These include effects of water waves on motion-blur, rolling shutter distortion and random tilting. They also include multi-view calibration and geometric recovery in a random environment, using limited embedded resources. Moreover, to cover large areas on Earth, we look into scalability. In conjunction to computer vision, solutions include a new design of drifting floats. We demonstrate a prototype network having two floating nodes, in real sea experiments, yielding information of cloud-base height, with no noticeable rolling shutter or motion blur artifacts.\nM.Sc. student under the supervision of Prof. Yoav Schechner.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:1061 Meyer Building &amp; Zoom&nbsp;
UID:eventx6a5a287eecc5610793
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20250506T143000
DTEND;TZID=Asia/Jerusalem:20250506T153000
DTSTAMP;TZID=Asia/Jerusalem:20250506T143000
SUMMARY: colloq  talk by Prof.  Yishai Mansour (TAU)  Probably Approximately Precision and Recall Learning  at 2025-05-06 14:30:00
DESCRIPTION:We introduce a Probably Approximately Correct (PAC) learning framework where each hypothesis is represented by a graph, with edges indicating positive interactions, such as between users and items. This framework subsumes the classical binary and multi-class PAC learning models as well as multi-label learning with partial feedback, where only a single random correct label per example is observed, rather than all correct labels.\nOur work uncovers a rich statistical and algorithmic landscape, with nuanced boundaries on what can and cannot be learned. Notably, classical methods like Empirical Risk Minimization fail in this setting, even for simple hypothesis classes with only two hypotheses. To address these challenges, we develop novel algorithms that learn exclusively from positive data, effectively minimizing both precision and recall losses. Specifically, in the realizable setting, we design algorithms that achieve optimal sample complexity guarantees. In the agnostic case, we show that it is impossible to achieve additive error guarantees (i.e., additive regret)&mdash;as is standard in PAC learning&mdash;and instead obtain meaningful multiplicative approximations.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 337
UID:eventx6a5a287eecc6910789
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20250506T163000
DTEND;TZID=Asia/Jerusalem:20250506T183000
DTSTAMP;TZID=Asia/Jerusalem:20250506T163000
SUMMARY: CSpecial Event  Special Lecture at the Faculty - Brigadier General (res.) Itai Brun, former Head of IDF's Intelligence Analysis Division  at 2025-05-06 16:30:00
DESCRIPTION:The Faculty of Computer Science at the Technion invites you to a special lecture:Back to the National Security Rooms After 7/10: Understanding Reality in the Post-Truth EraTuesday, May 6 at 4:30 PM Auditorium Taub 2, Faculty of Computer Science, Technion\nThis lecture is a reflection on the challenges of understanding reality in the post-truth era, through the lens of the October 7 surprise.\nSpeaker: Brigadier General (res.) Itai Brun, former Head of IDF's Intelligence Analysis DivisionBrigadier General (res.) Itai Brun served as Head of the Analysis Division in the IDF Intelligence Directorate from 2011 to 2014, and was called back to the position for several months in 2024 during the war in Gaza.Following his retirement from the IDF, he served for several years as Deputy Director of the Institute for National Security Studies (INSS) in Tel Aviv.Itai has published a series of books and articles on intelligence, air power, and national security, and currently teaches courses on these subjects at Tel Aviv University, Reichman University, and the University of Haifa.\nRegister here\nThe event is open to the entire faculty and not just to hackathon participants - everyone is welcome!!
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Auditorium Taub 2
UID:eventx6a5a287eecc7a10791
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20250507T113000
DTEND;TZID=Asia/Jerusalem:20250507T123000
DTSTAMP;TZID=Asia/Jerusalem:20250507T113000
SUMMARY: ceClub  talk by Yotam Gafni (Weizmann)  CE-Club - Designing Blockchain Fees  at 2025-05-07 11:30:00
DESCRIPTION:Miner fees are a key component in the incentive scheme of properly running Blockchains&rsquo; decentralized transaction processing. In determining how fees work, we should consider many different goals: Efficient allocation, simplicity for users, and robustness to possible manipulation vectors. We consider this problem through the lens of auction theory, and characterize the tradeoffs different mechanisms may offer, in particular w.r.t. the threat of miner-user collusion.The talk is based on joint works with Aviv Yaish, Matheus Ferreira, and Max Resnick.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Mayer 1061 &amp; Zoom
UID:eventx6a5a287eecc8e10794
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20250507T130000
DTEND;TZID=Asia/Jerusalem:20250507T140000
DTSTAMP;TZID=Asia/Jerusalem:20250507T130000
SUMMARY: Theory Semina  talk by Tom Waknine (Technion)  Theory Seminar - On Reductions of Learning Problems and the Borsuk-Ulam theorem  at 2025-05-07 13:00:00
DESCRIPTION:Many practical prediction algorithms represent inputs in Euclidean space and replace the discrete 0/1 classification loss with a real-valued surrogate loss, effectively reducing classification tasks to stochastic optimization. In this talk, I will explore the expressiveness of such reductions in terms of key resources. I will formally define a general notion of reductions between learning problems, and realte it some well known examples such as representations by half-spaces.I will then establish bounds on the minimum Euclidean dimension D required to reduce a concept class with VC dimension d to a Stochastic Convex Optimization (SCO) problem in a D dimensional Euclidean space. This result provides a formal framework for understanding the intuitive interpretation of the VC dimension as the number of parameters needed to learn a given class.The proof leverages a clever application of the Borsuk-Ulam theorem, illustrating an intersting connection between topology and learning theory.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 201
UID:eventx6a5a287eecc9f10795
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20250507T173000
DTEND;TZID=Asia/Jerusalem:20250507T193000
DTSTAMP;TZID=Asia/Jerusalem:20250507T173000
SUMMARY: CSpecial Event  Registration for Technology Workshops - Hackathon Preparation Week - Open to Everyone!  at 2025-05-07 17:30:00
DESCRIPTION:The CS Doing Good Faculty Hackathon is underway and we are excited!Even if you are not participating in the hackathon - that does not mean you cannot join the celebration: we have opened the preparation workshops to all students in the faculty!\nFeel like you have no experience?In preparation for the hackathon, we have created three workshops that will give you practical tools in AI development, Web development and an introduction to NLP.\nAI &amp; API4.5 at 17:30 on Taub 9Transitioning from a student's mindset to a hacker's mindset, soft and technical skills for hackathons, examples of using an external API and a technical workshop.\nWeb Development5.5 at 17:30 on Taub 9We will learn about web development - frontend, backend and what is in between. In the workshop we will learn the principles and basics of web development, including tools that will help you in the hackathon. In addition, you will build a small, high-quality fullstack application.\nNLP Workshop7.5 at 17:30 in Taub 9We will learn basic principles in NLP and working with libraries, including experimentation. No prior knowledge of NLP is required.\nRegistration is open to all students (even those who are not participating in the hackathon!) | You can register for more than one workshop | Registration in the form here\nWe are waiting for you!CS HACK Team
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:
UID:eventx6a5a287eeccb110792
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20250514T123000
DTEND;TZID=Asia/Jerusalem:20250514T143000
DTSTAMP;TZID=Asia/Jerusalem:20250514T123000
SUMMARY: CSpecial Event  CS Research Day 2025  at 2025-05-14 12:30:00
DESCRIPTION:The CS Research Day for graduate studies will be held on Wednesday, May 14, 2025 between 12:30-14:30, at the lobby of the CS Taub Building.\nResearch Day events are opportunity for our graduate students to expose their researches using posters and presentations to CS faculty and all degrees students, Technion distinguished representatives and to high-ranking delegates from the hi-tech leading industry companies in Israel and abroad.\nThe participating researches will be on various topics: Cryptology and Cyber, Data Centers and Clouds, Graphics, Intelligent Systems and Scientific Computation, Machine Learning and Information Retrieval, Systems and Applications, Testing and Verification, Theory of Computer Science.\nThe presenting researches
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:CS Taub Lobby
UID:eventx6a5a287eeccc610766
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20250514T130000
DTEND;TZID=Asia/Jerusalem:20250514T140000
DTSTAMP;TZID=Asia/Jerusalem:20250514T130000
SUMMARY: Theory Semina  talk by Niv Buchbinder (Tel-Aviv University)  Theory Seminar: Deterministic Algorithms and Faster Algorithms for Submodular Maximization subject to a Matroid Constraint  at 2025-05-14 13:00:00
DESCRIPTION:Maximization of submodular functions under various constraints is a fundamental problem that has been studied extensively.In this talk I will discuss submodular functions and interesting research questions in the field.I will present several new techniques that lead to both deterministic algorithms and faster randomized algorithms for maximizing submodular functions.In particular, for monotone submodular functions subject to a matroid constraint we design a deterministic non-oblivious local search algorithm that has an approximation guarantee of 1 &ndash; 1/e &ndash; \eps (for any \eps &gt; 0), vastly improving over the previous state-of-the-art 0.5008-approximation.For general (non-monotone) submodular functions we introduce a new tool, that we refer to as the extended multilinear extension, designed to derandomize submodular maximization algorithms that are based on the successful &ldquo;solve fractionally and then round&rdquo; approach.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 201
UID:eventx6a5a287eeccda10796
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20250518T133000
DTEND;TZID=Asia/Jerusalem:20250518T143000
DTSTAMP;TZID=Asia/Jerusalem:20250518T133000
SUMMARY: MSC  talk by Andrew Elashkin  Counterfactual and Robustness-Based Explanations for Reinforcement Learning Policies  at 2025-05-18 13:30:00
DESCRIPTION:Reinforcement learning policies in Markov decision processes (MDPs) often behave unexpectedly, especially in environments with sparse rewards, raising challenges for debugging and verification. We propose a general framework for discrete MDPs to generate two complementary, one-step explanations for single-action anomalies: (1) minimal counterfactual states&mdash;the smallest factored-state perturbations that flip a chosen action&mdash;and (2) robustness regions&mdash;contiguous state neighborhoods over which the original action remains invariant.\nWithout accessing internal model details, our black-box technique uses only action feedback, is applicable to any discrete RL setting, and has been validated on various Gymnasium environments, providing actionable understanding of when and why policies change their decisions.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 301 &amp; Zoom
UID:eventx6a5a287eeccec10797
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20250521T113000
DTEND;TZID=Asia/Jerusalem:20250521T123000
DTSTAMP;TZID=Asia/Jerusalem:20250521T113000
SUMMARY: ceClub  talk by Nicolás Wainstein, PhD (Research Fellow | AMSG | Faculty of Electrical and Computer Engineering Technion – Israel Institute of Technology)  CE-Club - Towards Energy-Efficient AI Hardware: Mixed-Signal In-Memory Computing and Ultra-Dense Die-to-Die Links  at 2025-05-21 11:30:00
DESCRIPTION:The rapid advancement of artificial intelligence (AI) is pushing the limits of conventional digital computing architectures. Key performance bottlenecks&ndash;computational efficiency, memory bandwidth, and interconnect performance&ndash;are becoming increasingly critical, especially for edge AI applications that demand low latency and ultra-low power consumption. In this talk, I will present my recent research aimed at overcoming these challenges through a multidisciplinary approach focusing on three main pillars: 1) analog/mixed-signal circuits, 2) in-memory computing (IMC), and 3) high-speed die-to-die (D2D) links. I will show how time-domain IMC using nonvolatile emerging memory technologies, such as ferroelectric FETs, can improve the energy-efficiency of AI hardware, as demonstrated by a prototype in 28 nm CMOS. I will also present a mixed-mode fast-locking delay-locked loop for latency-critical parallel links (such as D2D), implemented in a 3 nm FinFET CMOS process. These works span CMOS and emerging device technologies, combining insights from device physics, circuit design, and system architecture to enable the next generation of high-performance, energy-efficient AI hardware.\nBio: Nicolas Wainstein is a Research Fellow at the ECE faculty at the Technion. From 2021 to 2024, he was a Senior Analog/Mixed-Signal Design Engineer and Technical Lead at Intel, Israel, working on high-speed parallel wireline links, such as DDR and die-to-die (D2D) communication. He earned his PhD in Electrical Engineering from the Technion, supervised by Prof. Shahar Kvatinsky and Prof. Eilam Yalon. Nicol&aacute;s was the recipient of several awards, including the Hershel Rich Innovation Award, the IEEE Electron Devices Society Ph.D. Student Fellowship (Europe and Middle East region), the Yablonovitch Research Prize, the 1st place in the RBNI Prize for Excellence in Nanoscience and Nanotechnology, and the Jury Award for Outstanding Students.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Mayer 1061 &amp; Zoom
UID:eventx6a5a287eecd0e10798
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20250521T130000
DTEND;TZID=Asia/Jerusalem:20250521T140000
DTSTAMP;TZID=Asia/Jerusalem:20250521T130000
SUMMARY: Theory Semina  talk by Yaroslav Alekseev (Technion)  Theory Seminar: Lower Bounds For Bounded-Depth Resolution Over Parities  at 2025-05-21 13:00:00
DESCRIPTION:In this talk, I will explain the basics of proof complexity, its connections to NP vs coNP, and SAT solvers lower bounds. One of the frontier proof systems, for which we do not have any lower bounds, is the Res(+) proof system. A recent breakthrough by Efremenko, Garlik, and Itsykson (STOC 2024) established an exponential lower bound for regular Res(+). In our recent work, we proved an exponential lower bound for bounded-depth Res(+).\nWe will discuss the methods used to prove those lower bounds, such as lifting theorems and Prover-Delayer games.&nbsp;
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 201
UID:eventx6a5a287eecd2c10804
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20250525T143000
DTEND;TZID=Asia/Jerusalem:20250525T153000
DTSTAMP;TZID=Asia/Jerusalem:20250525T143000
SUMMARY: MSC  talk by Shoham Shimon Berrebi  Labeled Coupons' Collector Problem  at 2025-05-25 14:30:00
DESCRIPTION:We extend the Coupon Collector's Problem (CCP) and present a novel generalized model, referred as the \ekecd problem, where one is interested in recovering a bipartite graph with a perfect matching, which represents the coupons and their matching labels. We show two extra-extensions to this variation: the heterogeneous sample size case (\efmecd) and the partly recovering case.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Room 601 &amp; Zoom
UID:eventx6a5a287eecd3c10799
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20250526T113000
DTEND;TZID=Asia/Jerusalem:20250526T123000
DTSTAMP;TZID=Asia/Jerusalem:20250526T113000
SUMMARY: MSC  talk by Adi Harif  Ductape: Optimizing Dynamically Typed Programs using Ahead-of-Time Compilation and Data-Flow Analysis  at 2025-05-26 11:30:00
DESCRIPTION:The software landscape is showing consistent, accelerated growth in the volume of code developed using dynamic languages. These languages are characterized by dynamic typing and more lax preemptive checks, which allow rapid application development and shorter development cycles. The vast majority of these languages are interpreted; that is, programs are executed by an interpreter in a managed runtime environment. Such interpreters incur significant performance hits, and, to counter that, modern runtime environments usually employ some sort of optimization. The most common one is Just-in-Time compilation (JIT), which translates source code on-demand into native code that can run much faster. Some notable JIT engines (such as V8 for JavaScript) exhibit impressive speedups. Still, in most realistic scenarios, they cannot surpass the performance of hand-crafted native code written in a low-level language like C.\nThere are inherent reasons for why Ahead-of-Time compilation (AOT) is rarely practiced with dynamic languages. Since variables are dynamically typed, this will require most of the type-checking to be done at runtime still, thus limiting the range of optimization that can be performed ahead of time, consequently limiting the benefit of AOT compilation. We propose an approach that utilizes static analysis for the purpose of sound type inference, which can then be leveraged for code generation requiring minimal amount of runtime type checks. Unlike previous work in this area, our approach eliminates the need for the JIT at runtime, or, indeed, any managed runtime at all. We claim that real-world JavaScript applications can, in fact, benefit much from using AOT, in terms of increased speed. We show empirical evidence on a set of representative benchmarks supports this claim.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 301
UID:eventx6a5a287eecd4d10801
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20250527T113000
DTEND;TZID=Asia/Jerusalem:20250527T133000
DTSTAMP;TZID=Asia/Jerusalem:20250527T113000
SUMMARY: pixel-club  talk by Inbal Kom Betzer  Pixel Club - Learning-Based Polarized Scattering Tomography of Clouds  at 2025-05-27 11:30:00
DESCRIPTION:Inverse problems in scientific imaging often seek physical characterization of heterogeneous scene materials. The scene is thus represented by physical quantities, such as the density and sizes of particles (microphysics) across a domain. Moreover, the forward image formation model is physical. An important case is that of clouds, where microphysics in three dimensions (3D) dictate the cloud dynamics, lifetime and albedo, with implications to Earth&rsquo;s energy balance, sustainable energy and rainfall. Current methods, however, recover very degenerate representations of microphysics. To enable 3D volumetric recovery of all the required microphysical parameters, we introduce the neural microphysics field (NeMF). It is based on a deep neural network, whose input is multi-view polarization images. NeMF is pre-trained through supervised learning. Training relies on polarized radiative transfer, and noise modeling in polarization-sensitive sensors. The results offer unprecedented recovery, including droplet effective variance. We test NeMF in rigorous simulations and demonstrate it using real-world polarization-image data.\nM.Sc. student under the supervision of Prof. Yoav Schechner.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom &amp; 506, Zisapel Building
UID:eventx6a5a287eecd6110809
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20250528T130000
DTEND;TZID=Asia/Jerusalem:20250528T140000
DTSTAMP;TZID=Asia/Jerusalem:20250528T130000
SUMMARY: MSC  talk by Guy Arbel  Determinization of Min-Plus Weighted Automata  at 2025-05-28 13:00:00
DESCRIPTION:Min-Plus Weighted Finite Automata (WFAs) are a quantitative extension of Boolean automata whereby each word is assigned an integer, instead of being accepted or rejected.\nApplications of WFAs fall on a wide spectrum including verification, rewriting systems, tropical algebra, speech and image processing and have been key to proving the star-height conjecture.&nbsp;Unlike Boolean automata, WFAs cannot always be determinized. The decidability of whether a WFA admits an equivalent deterministic WFA is a long standing open problem.\nWe prove that this problem is decidable.As part of the proof, we develop a new toolbox for reasoning about the run structure of weighted automata.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 601 &amp; Zoom
UID:eventx6a5a287eecd7410800
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20250528T130000
DTEND;TZID=Asia/Jerusalem:20250528T140000
DTSTAMP;TZID=Asia/Jerusalem:20250528T130000
SUMMARY: Theory Semina  talk by Noa Schiller (Tel-Aviv University)  Theory Seminar: History-Independent Concurrent Hash Tables  at 2025-05-28 13:00:00
DESCRIPTION:A history-independent data structure does not reveal the history of operations applied to it, only its current logical state, even if its internal state is examined. While history independence has been extensively studied in sequential data structures, until very recently, it was not studied in a concurrent setting, where the data structure can be accessed by many threads at the same time.\nIn this talk I will discuss history-independent concurrent dictionaries, in particular, hash tables, and establish inherent bounds on their space requirements. We show that there is a lock-free history-independent concurrent hash table, in which each memory cell stores two elements and two bits, based on Robin Hood hashing. The expected amortized step complexity of the hash table is O(c), where c is an upper bound on the number of concurrent operations that access the same element, assuming the hash table is not overpopulated.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 201
UID:eventx6a5a287eecd8510810
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20250528T193000
DTEND;TZID=Asia/Jerusalem:20250528T220000
DTSTAMP;TZID=Asia/Jerusalem:20250528T193000
SUMMARY: CSpecial Event  Computer Science Faculty Bands Evening  at 2025-05-28 19:30:00
DESCRIPTION:You are invited to the Computer Science Faculty Bands Evening!!\nTalented students and faculty members from our faculty in a variety of ensembles and musical styles.Join us for an unforgettable evening of live performances, a break from the stress and studies, and incredible energy.\nWednesday, May 28th at 7:30 PM on the Taub Terrace\nSo come on, get your strings ready, admission is free!
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub Terrace
UID:eventx6a5a287eecd9510807
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20250529T110000
DTEND;TZID=Asia/Jerusalem:20250530T140000
DTSTAMP;TZID=Asia/Jerusalem:20250529T110000
SUMMARY: CSpecial Event  Open Day for those interested in studying at the Technion – Thursday and Friday, May 29-30, in Sarona, Tel Aviv  at 2025-05-29 11:00:00
DESCRIPTION:Open Day for those interested in studying at the Technion &ndash; Thursday and Friday, May 29-30, in Sharona, Tel Aviv.\n- Thursday, May 29 between 11:00-18:00 &ndash; An opportunity to deepen and get to know the study paths at the faculty, meet the staff and best prepare for the beginning of the journey\n- Friday, May 30 between 10:00-14:00 &ndash; Invitation to coffee with faculty graduates, students and staff.\nTo register for the open day: here
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Sarona, Tel Aviv
UID:eventx6a5a287eecda610805
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20250603T100000
DTEND;TZID=Asia/Jerusalem:20250603T110000
DTSTAMP;TZID=Asia/Jerusalem:20250603T100000
SUMMARY: MSC  talk by Roi Yona  Using Database Dependencies to Constrain Approval-Based Committee Voting in the Presence of Context  at 2025-06-03 10:00:00
DESCRIPTION:In Approval-Based Committee (ABC) voting, &nbsp;each voter lists the candidates they approve and then a voting rule aggregates the individual approvals into a committee that represents the collective choice of the voters.&nbsp;An extensively studied class of such rules is the class of ABC scoring rules, where each voter contributes to each possible committee a score based on the voter's approvals.&nbsp;We initiate a study of ABC voting in the presence of constraints about the general context surrounding the candidates.&nbsp;Specifically, we consider a framework in which there is a relational database with information about the candidates together with integrity constraints on the relational database extended with a virtual relation representing the committee.&nbsp;For an ABC scoring rule, the goal is to find a committee of maximum score such that all integrity constraints hold in the extended database.&nbsp;\nWe focus on two well-known types of integrity constraints in relational databases:&nbsp;tuple-generating dependencies (TGDs) and denial constraints (DCs).&nbsp;The former can express, for example, desired representations of groups, while the latter can express conflicts among candidates. &nbsp;ABC voting is known to be computationally hard without integrity constraints, except for the case of approval voting where it is tractable.&nbsp;We show that integrity constraints make the problem NP-hard for approval voting, but we also identify certain tractable cases when key constraints are used.&nbsp;We then present an implementation of the framework via a reduction to Mixed Integer Programming (MIP) that supports arbitrary ABC scoring rules, TGDs and DCs.&nbsp;We devise heuristics for optimizing the resulting MIP, and describe an empirical study that illustrates the effectiveness of the optimized MIP over databases in three different domains.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 401
UID:eventx6a5a287eecdb710803
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20250604T130000
DTEND;TZID=Asia/Jerusalem:20250604T140000
DTSTAMP;TZID=Asia/Jerusalem:20250604T130000
SUMMARY: Theory Semina  talk by Shay Golan (Reichman University and Haifa University)  Theory Seminar: The Complexity of Dynamic LZ77 is Θ~(n^{2/3})  at 2025-06-04 13:00:00
DESCRIPTION:The Lempel-Ziv 77 (LZ77) factorization is a fundamental compression scheme widely used in text processing and data compression. In this work, we investigate the time complexity of maintaining the LZ77 factorization of a dynamic string. By establishing matching upper and lower bounds, we fully characterize the complexity of this problem. &nbsp;\nWe present an algorithm that efficiently maintains the LZ77 factorization of a string $S$ undergoing edit operations, including character substitutions, insertions, and deletions. Our data structure can be constructed in $tilde{O}(n)$ time for an initial string of length $n$ and supports updates in $tilde{O}(n^{2/3})$ time, where $n$ is the current length of $S$. Additionally, we prove that no algorithm can achieve an update time of $O(n^{2/3-varepsilon})$ unless the Strong Exponential Time Hypothesis fails. This lower bound holds even in the restricted setting where only substitutions are allowed and only the length of the LZ77 factorization is maintained.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 201
UID:eventx6a5a287eecdcb10813
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20250608T093000
DTEND;TZID=Asia/Jerusalem:20250608T160000
DTSTAMP;TZID=Asia/Jerusalem:20250608T093000
SUMMARY: CSpecial Event  Celebrating the Honorary Doctorate of  Prof. Maurice Herlihy Seminar  at 2025-06-08 09:30:00
DESCRIPTION:You are invited to a seminar to celebrating the honorary doctorate to Professor Maurice Herlihy, which will be held on Sunday, June 8, 2025, in the Taub Building.\nSeminar schedule:\n9:30-10:30 Piano Auditorium (chair: Prof. Hagit Attiya, Technion)* Greetings: Prof. Danny Raz (Dean, Computer Science, Technion)* Prof. Sergio Rajsbuam (UNAM): The intimate relationship between distributed computing and topology\n10:30-11:00 coffee break\n11:00-13:15 Piano Auditorium (chair: Prof. Yoram Moses)* Prof. Yossi Gilad (Hebrew University): Scaling Byzantine Agreement with Algorand* Prof. Naama Ben-David (Technion): Byzantine Agreement with Predictions* Prof. Adam Morrison (Tel-Aviv University): Occualizer: Optimistic Concurrent * Search Trees From Sequential* Prof. Erez Petrank (Technion): Memory Reclamation for Concurrent Data Structures&nbsp;Code* Prof. Hagit Attiya (Technion): A gold standard, but not a Silver Bullet\n13:15-14:30 light lunch\n14:30-16:00 Taub 2 (chair: Prof. Erez Petrank, Technion)* Greetings: Prof. Idit Keidar (Dean, Electrical and Computer Engineering, Technion)* Prof. Maurice Herlihy (Brown University): Cross-Chain Consensus\nTo register, follow the link
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:
UID:eventx6a5a287eecdda10806
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20250608T143000
DTEND;TZID=Asia/Jerusalem:20250608T160000
DTSTAMP;TZID=Asia/Jerusalem:20250608T143000
SUMMARY: colloq  talk by Maurice Herlihy (Brown University)  Cross-Chain Consensus  at 2025-06-08 14:30:00
DESCRIPTION:As decentralized ledgers and blockchains become more common, &nbsp;cross-chain interoperability becomes essential to making those ledgers useful. In a cross-chain task, $m$ active, Byzantine parties undertake to trade assets using n passive but trustworthy smart contracts. Each party seeks an outcome that maximizes its own utility in the presence of Byzantine counterparties. This talk introduces a novel task called ``cross-chain consensus''.We show that cross-chain consensus is \emph{universal}, meaning that any cross-chain consensus protocol can be transformed into a protocol for any other well-formed cross-chain task. We show that cross-chain consensus is impossible using unsigned messages, even if communication channels are authenticated. If each party can generate signed messages verifiable on all the blockchains, then there is a tight bound of $\Theta(m)$ communication rounds, where $m$ is the number of participating parties. Moreover, there is a communication-round optimal protocol that uses a finite number of precomputed signed messages.Joint work with Sucharita Jayanti.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 2
UID:eventx6a5a287eecdeb10811
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20250611T120000
DTEND;TZID=Asia/Jerusalem:20250611T130000
DTSTAMP;TZID=Asia/Jerusalem:20250611T120000
SUMMARY: MSC  talk by Shani Goren  Hierarchical Selective Classification  at 2025-06-11 12:00:00
DESCRIPTION:In this seminar I will present our paper that was published in NeurIPS 2024. We introduce hierarchical selective classification, which extends selective classification to a hierarchical setting. Our approach leverages the inherent structure of class relationships, enabling models to reduce the specificity of their predictions when faced with uncertainty. We formalize hierarchical risk and coverage, and introduce hierarchical risk-coverage curves. Next, we develop algorithms for hierarchical selective classification (which we refer to as "inference rules"), and propose an efficient algorithm that guarantees a target accuracy constraint with high probability. Lastly, we conduct extensive empirical studies on over a thousand ImageNet classifiers, revealing that training regimes such as CLIP, pretraining on ImageNet21k and knowledge distillation boost hierarchical selective performance.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom
UID:eventx6a5a287eecdfc10802
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20250611T130000
DTEND;TZID=Asia/Jerusalem:20250611T140000
DTSTAMP;TZID=Asia/Jerusalem:20250611T130000
SUMMARY: Theory Semina  talk by Tomer Even (Technion)  Theory Seminar: Output-Sensitive Approximate Counting Via A Measure-Bounded Hyperedge Oracle. Or: How Asymmetry Helps Estimate K-Clique Counts Faster  at 2025-06-11 13:00:00
DESCRIPTION:Detecting a k-clique in a graph, for k at least 3, is a fundamental problem in fine-grained complexity, conjectured to require n^(omega * k / 3 - o(1)) time, where omega is the matrix multiplication exponent. In this talk, we present new detection and approximate counting algorithms for graphs containing many k-cliques.\nThe talk is based on a joint work with Keren Censor-Hillel and Virginia Vassilevska Williams. To appear in STOC 2025.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 201
UID:eventx6a5a287eece0c10815
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20250618T110000
DTEND;TZID=Asia/Jerusalem:20250618T120000
DTSTAMP;TZID=Asia/Jerusalem:20250618T110000
SUMMARY: MSC  talk by Oren Afek  PaLaDiN - Time-travel Debugging with Semantic Queries  at 2025-06-18 11:00:00
DESCRIPTION:The process of debugging software can be a time-consuming and tedious task, often requiring developers to guess which lines of code to set breakpoints on in order to gather more information. Time-travel debuggers are tools that record the program's state throughout its run and let the user inspect it afterwards, a.k.a "Post-Mortem". While time travel debuggers have reduced the time required to re-run a program, they still require developers to view the program on a linear timeline, jumping through different time points. To address these challenges, we have developed a new debugging tool called PaLaDiN. PaLaDiN offers several advantages over traditional debugging methods, including a concise, high-level yet expressive query DSL for retrieving trace data, the ability to show summary information for entire program sections, and the ability to focus on a specific slice of the program's run by running high-level predicates. PaLaDiN also allows developers to connect different, possibly unrelated, fragments of the program using join operators, run queries with alternative states, analyze variable lifetimes throughout the program's run, and compare different implementations of the same function. Overall, PaLaDiN provides a comprehensive and flexible approach to debugging that can significantly reduce the time and effort required to identify and fix software bugs.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom
UID:eventx6a5a287eece1b10814
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20250622T093000
DTEND;TZID=Asia/Jerusalem:20250622T103000
DTSTAMP;TZID=Asia/Jerusalem:20250622T093000
SUMMARY: PHD  talk by Barak Pinkovich  A Multi-Resolution Probabilistic Approach to Urban Aerial Exploration  at 2025-06-22 09:30:00
DESCRIPTION:This doctoral research focuses on advancing the autonomy of Unmanned Aerial Vehicles (UAVs) operating in complex, dense urban environments, particularly emphasizing the critical task of identifying safe landing locations. Traditional sensing and decision-making methods often struggle with the intricacies of such settings. This work proposes leveraging semantic segmentation, a machine-learning technique that interprets visual scenes by assigning a class label (e.g., road, building, vegetation) to each pixel in an image, as a sophisticated sensor modality for UAVs.\nA primary contribution is a novel multi-resolution strategy for landing site selection. This method involves the UAV collecting visual information at progressively decreasing altitudes, thereby acquiring images with increasing spatial resolution. For each potential terrain patch, a probability distribution representing its suitability for landing is maintained and updated as new data from different altitudes becomes available. A landing site is confirmed once the confidence level for a patch surpasses a predefined threshold. This decision-making process relies on a semantic segmentation algorithm to provide per-pixel likelihoods of different terrain types, which are then integrated with prior knowledge and previous measurements.\nTo facilitate the development and rigorous evaluation of semantic segmentation algorithms tailored for drone applications, a comprehensive dataset named MESSI (Multi-Elevation Semantic Segmentation Image) was developed. MESSI is distinguished by its inclusion of images captured from a wide range of altitudes and diverse urban locations, reflecting the varied perspectives encountered during a realistic 3D drone flight. This richly annotated dataset serves as a valuable resource for training deep neural networks and benchmarking their performance in understanding urban scenes.\nFurthermore, this research addresses a fundamental challenge in applying semantic segmentation to probabilistic search problems: the statistical nature of its outputs differs from assumptions in classical detection theory. A systematic methodology is introduced to bridge this gap, enabling the effective integration of semantic segmentation into established probabilistic search frameworks. This integration provides a more robust foundation for decision-making processes, such as the identification of viable landing sites in partially known environments. The practical feasibility and effectiveness of the proposed approaches are demonstrated through extensive simulations and validation using real-world datasets, highlighting their potential to significantly enhance UAV operational capabilities in challenging urban settings.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom
UID:eventx6a5a287eece2d10812
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20250626T143000
DTEND;TZID=Asia/Jerusalem:20250626T153000
DTSTAMP;TZID=Asia/Jerusalem:20250626T143000
SUMMARY: MSC  talk by Maymana Hasan  Deep Learning Approach for Studying mRNA Translation Regulation  at 2025-06-26 14:30:00
DESCRIPTION:Regulation of mRNA translation, particularly under stress conditions, is a critical layer of gene expression control. Many of the elements that regulate translation during stress are embedded in the sequence of the mRNA, however, decoding how sequence regulates translation is still a largely unresolved question. Here we decided to harness the power of deep learning, in order to try and identify sequences that actively regulate translation, in particular, uORFs (upstream Open Readin Frames).\nIn this seminar, I will present our deep learning-based approach that utilizes experimental data from ribosome profiling (ribo-seq) experiments, to predict actively-regulating uORFs from different types. By augmentation of the training data using a new approach to generate a large synthetic database for pre-training, our model achieved improved classification performance of different uORF subtypes.\nThis approach demonstrates the potential of deep learning to advance our understanding of translation regulation.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom
UID:eventx6a5a287eece4310817
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20250626T143000
DTEND;TZID=Asia/Jerusalem:20250626T153000
DTSTAMP;TZID=Asia/Jerusalem:20250626T143000
SUMMARY: MSC  talk by Priel Hazan  Automating ECG Interpretation: Balanced Classification For Rare Diseases And Signal Digitization  at 2025-06-26 14:30:00
DESCRIPTION:Standard 12-lead ECGs provide a multi-perspective view of cardiac electrophysiology by capturing 12 signals that encode critical information on the propagation of cardiac electrical activity. However, training models on multilabel ECG datasets presents a major challenge: the inherent imbalance between common and rare diseases leads to suboptimal performance, particularly on rare pathologies. One approach to mitigating class imbalance is to over-sample instances associated with rare diseases. While balanced-sampling is straightforward in multiclass settings, sampling becomes more complex in multilabel classification due to the presence of multiple labels per data point.\nTo address this, this work proposes a learnable sampling framework for multilabel classification. Unlike heuristic samplers in the literature that rely on discrete instance duplication, this approach formulates sampling as a continuous constrained optimization problem, with the solution being a probability distribution over instances that guides the training sampling process. The framework dynamically adjusts sampling probabilities to ensure balanced exposure across all disease classes. The method integrates class-specific prioritization directly into the optimization objective, giving clinicians the flexibility to focus training on critical (e.g., life-threatening) conditions.\nAnother pillar of ECG automation is the digitization of printed records. While modern ECG machines produce digital signals, patients often receive printed records, limiting the applicability of automated ECG interpretation and ECG-based decision support systems. Digitization, the process of extracting signals from printed records, is challenging due to variations in rotation, paper formats, layouts, and inconsistent lighting in mobile-captured images.\nTo address these challenges, this work proposes DigiECG: a format-agnostic pipeline that corrects rotation, detects leads, extract signals using a segmentation model, and restores the time scale. The proposed pipeline achieves state-of-the-art performance compared to recent ECG digitization methods, with a Root Mean Squared Error (RMSE) of 52.3 microvolts and a rotation correction RMSE of 0.71 degrees.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom&nbsp;
UID:eventx6a5a287eece5510819
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20250702T130000
DTEND;TZID=Asia/Jerusalem:20250702T150000
DTSTAMP;TZID=Asia/Jerusalem:20250702T130000
SUMMARY: CSpecial Event  Closing The Year With Google - Final Spotlight Day For 2025!  at 2025-07-02 13:00:00
DESCRIPTION:Closing the year with Google coming to the faculty for a final spotlight day for 2025!\nWednesday, 2.7.2025 | 13:00 | Taub Auditorium 2\nWhat's on the program:Want to discover how billions of users are connected? GoogleJoin a special event with Google engineers that will include an exclusive technology interview workshop with tips, tricks and behind-the-scenes secrets that will help you get accepted.\nPlus: career paths, mingling, Q&amp;A and of course delicious refreshments\nThe number of places is limited - details and registration here
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 2 Auditorium&nbsp;
UID:eventx6a5a287eece6910822
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20250702T130000
DTEND;TZID=Asia/Jerusalem:20250702T140000
DTSTAMP;TZID=Asia/Jerusalem:20250702T130000
SUMMARY: Theory Semina  talk by Itai Boneh (Reichman Univesity and the University of Haifa)  Theory Seminar: Õptimal Fault-Tolerant Labeling for Reachability and Approximate Distances in Directed Planar Graphs  at 2025-07-02 13:00:00
DESCRIPTION:We present a labeling scheme that assigns labels of size &Otilde;(1) to the vertices of a directed weighted planar graph G, such that for any fixed ϵ&gt;0 from the labels of any three vertices s, t and f one can determine in &Otilde;(1) time a (1+ϵ)-approximation of the s-to-t distance in the graph G \ {f}.\nFor approximate distance queries, prior to our work, no efficient solution existed, not even in the centralized oracle setting.\nEven for the easier case of reachability, &Otilde;(1) queries were known only with a centralized oracle of size &Otilde;(n) [SODA 21].\nJoint work with Shiri Chechik, Shay Golan, Shay Mozes, and Oren Weimann.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 201
UID:eventx6a5a287eece7910823
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20250703T153000
DTEND;TZID=Asia/Jerusalem:20250703T163000
DTSTAMP;TZID=Asia/Jerusalem:20250703T153000
SUMMARY: MSC  talk by Noor Athamnah  Linear Prover IOPs in Log Star Rounds  at 2025-07-03 15:30:00
DESCRIPTION:Interactive Oracle Proofs (IOPs) form the backbone of some of the most efficient general-purpose cryptographic proof-systems. In an IOP, the prover can interact with the verifier over&nbsp;multiple rounds, where in each round the prover sends a long message, from which the verifier&nbsp;only queries a few symbols.\nState-of-the-art IOPs achieve a linear-size prover and a poly-logarithmic verifier but require a&nbsp;relatively large, logarithmic, number of rounds. While the Fiat-Shamir heuristic can be used to&nbsp;eliminate the need for actual interaction, in modern highly-parallelizable computer architectures&nbsp;such as GPUs, the large number of rounds translates into a major bottleneck for the prover&nbsp;(since it needs to alternate between computing the IOP messages and the Fiat-Shamir hashes).\nMotivated by this fact, in this work we study the round complexity of linear-prover IOPs.\nOur main result is an IOP for a large class of Boolean circuits, with only O(log*(S)) rounds,&nbsp;where log* denotes the iterated logarithm function (and S is the circuit size). The prover has&nbsp;linear size O(S) and the verifier runs in time polylog(S) and has query complexity O(log*(S)).\nThe protocol is both conceptually simpler, and strictly more efficient, than prior linear prover&nbsp;IOPs for Boolean circuits.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom &amp; Taub 601
UID:eventx6a5a287eece8910821
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20250707T113000
DTEND;TZID=Asia/Jerusalem:20250707T123000
DTSTAMP;TZID=Asia/Jerusalem:20250707T113000
SUMMARY: MSC  talk by Shay Segal  Debugging into Existence with Program Synthesis  at 2025-07-07 11:30:00
DESCRIPTION:When modifying an existing codebase to handle new functionality, programmers will often debug the program until the insertion point for the new code.This method, termed Debugging into Existence, helps programmers familiarize themselves with the surrounding code and runtime state. Despite its real-world usage, it is limited by the inability to test potential code past the first time the location is called, since added functionality would change the future state making it irrelevant.\nPrior work has pioneered Live Execution over partial programs, with extensions using the provided values for synthesis by Programming by Example. In this work, we present DeSynt, a debugger extension that integrates live execution and program synthesis to extend the Debugging into Existence interaction model. DeSynt grants programmers meaningful run-time information across many executions, by allowing them to manipulate program state according to the desired functionality. Based on the state provided by the programmer, DeSynt then synthesizes programs that capture this functionality. We evaluated DeSynt in a between-subjects study on 10 users, and found that in tasks that do not involve complex fault localization, DeSynt reduces time to completion and concentrates programmer effort into fewer code locations. In addition, we found that users that used DeSynt spent more of their task time debugging, indicating DeSynt supports Debugging into Existence for those that already use it.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Room 601
UID:eventx6a5a287eece9c10818
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20250707T123000
DTEND;TZID=Asia/Jerusalem:20250707T133000
DTSTAMP;TZID=Asia/Jerusalem:20250707T123000
SUMMARY: PHD  talk by Hadas Orgad  Explaining, Improving and Evaluating Robustness of AI Models  at 2025-07-07 12:30:00
DESCRIPTION:Artificial Intelligence (AI), particularly neural networks, has become central to a wide array of applications &mdash; from language modeling to text-to-image generation. Despite these achievements, ensuring the robustness of AI models remains a significant challenge. Robustness refers to the ability of models to maintain performance across diverse inputs and avoid issues such as out-of-distribution failures, generation of harmful or incorrect content, and the propagation of social biases. Addressing robustness is crucial for deploying reliable AI systems in real-world scenarios.\nMotivated by these challenges, this thesis aims to improve the understanding, evaluation, and ultimately the robustness of AI models through interpretability-based methods. Interpretability research, which aims to elucidate the decision-making processes of these models, offers a promising pathway to address robustness challenges with customizable and cost-effective methods. In this seminar, I will present our research on enhancing AI robustness by applying insights from interpretability studies, focusing on mitigating biases, reducing harmful content, improving adaptability, and addressing hallucinations.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 301
UID:eventx6a5a287eecead10825
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20250707T183000
DTEND;TZID=Asia/Jerusalem:20250707T193000
DTSTAMP;TZID=Asia/Jerusalem:20250707T183000
SUMMARY: MSC  talk by Eyal Kushnir  Combinatorially Homomorphic Encryption  at 2025-07-07 18:30:00
DESCRIPTION:Homomorphic encryption enables public computation over encrypted data. In the past few decades, homomorphic encryption has become a staple of both the theory and practice of cryptography.\nNevertheless, while there is a general loose understanding of what it means for a scheme to be homomorphic, to date there is no single unifying minimal definition that captures all schemes. In this work, we propose a new definition, which we refer to as combinatorially homomorphic encryption, which attempts to give a broad base that captures the intuitive meaning of homomorphic encryption and draws a clear line between trivial and nontrivial homomorphism.\nOur notion relates the ability to accomplish some task when given a ciphertext, to accomplishing the same task without the ciphertext, in the context of communication complexity.\nThus, we say that a scheme is combinatorially homomorphic if there exists a communication&nbsp;complexity problem f(x, y) (where x is Alice&rsquo;s input and y is Bob&rsquo;s input) which requires communication c, but can be solved with communication less than c when Alice is given in addition&nbsp;also an encryption Ek(y) of Bob&rsquo;s input (using Bob&rsquo;s key k).\nWe show that this definition indeed captures pre-existing notions of homomorphic encryption&nbsp;and (suitable variants are) sufficiently strong to derive prior known implications of homomorphic&nbsp;encryption in a conceptually appealing way. These include constructions of (lossy) public-key encryption from homomorphic private-key encryption, as well as collision-resistant hash&nbsp;functions and private information retrieval schemes.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom&nbsp;
UID:eventx6a5a287eecebd10820
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20250709T110000
DTEND;TZID=Asia/Jerusalem:20250709T120000
DTSTAMP;TZID=Asia/Jerusalem:20250709T110000
SUMMARY: MSC  talk by Benyamin Trachtenberg  Strategic Classification with Non-Linear Classifiers  at 2025-07-09 11:00:00
DESCRIPTION:In strategic classification, the standard supervised learning setting is extended to support the notion of strategic user behavior in the form of costly feature manipulations made in response to a classifier. While standard learning supports a broad range of model classes, the study of strategic classification has, so far, been dedicated mostly to linear classifiers. This work aims to expand the horizon by exploring how strategic behavior manifests under non-linear classifiers and what this implies for learning. We take a bottom-up approach showing how non-linearity affects decision boundary points, classifier expressivity, and model classes complexity. A key finding is that universal approximators (e.g., neural nets) are no longer universal once the environment is strategic. We demonstrate empirically how this can create performance gaps even on an unrestricted model class.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 401 &amp; Zoom
UID:eventx6a5a287eecece10826
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20250709T123000
DTEND;TZID=Asia/Jerusalem:20250709T143000
DTSTAMP;TZID=Asia/Jerusalem:20250709T123000
SUMMARY: CSpecial Event  The Annual Project Fair Of The Taub Faculty Of Computer Science  at 2025-07-09 12:30:00
DESCRIPTION:We are pleased to invite you to the annual project fair of the Taub Faculty of Computer Science along with the Outstanding Project Competition - Wednesday, July 9, starting at 12:30 in the Taub Lobby - Floor 0.\nEveryone is invited to cheer on the competing teams and be impressed by the significant and creative projects.\nWe are waiting for you!
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub Lobby - Floor 0
UID:eventx6a5a287eecede10824
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20250709T130000
DTEND;TZID=Asia/Jerusalem:20250709T140000
DTSTAMP;TZID=Asia/Jerusalem:20250709T130000
SUMMARY: Theory Semina  talk by Ilan Komargodski (Hebrew University)  Theory Seminar: Proof-of-Useful-Work Blockchains  at 2025-07-09 13:00:00
DESCRIPTION:The recent emergence of Proof-of-Useful-Work (PoUW) consensus protocols [Komargodski,Schen,Weinstein'25] enable participants to reuse their *native* workloads (arbitrary matrix-multiplications) to secure blockchains as in Nakamoto's original Bitcoin protocol, allowing miners to earn external rewards. PoUW blockchains thereby give rise to a new economy where *both* data and compute are needed to efficiently mine the network.\nIn the first part of the talk, we will briefly describe the longstanding challenges of PoUW and how the protocol of [KSW'25] overcomes them. We then investigate and discuss the market dynamics and decentralization of PoUW cryptocurrencies. We extend the model of Fiat, Karlin, Koutsoupias and Papadimitrious (2019) to include external rewards, and analyze the resulting equilibrium. Our findings suggest that in some cases, miners with access to external incentives will optimize profitability by concentrating their useful tasks in a single block. We also point out initial directions for predicting the amount of useful-vs-nonuseful work done on chain, which dictates the energy savings of the KSW Blockchain.\nJoint works with Yogev Bar-On, Itamar Schen, and Omri Weinsteinhttps://arxiv.org/abs/2504.09971
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 201
UID:eventx6a5a287eeceeb10828
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20250716T103000
DTEND;TZID=Asia/Jerusalem:20250716T113000
DTSTAMP;TZID=Asia/Jerusalem:20250716T103000
SUMMARY: PHD  talk by Edo Dotan  Deep Learning for Biology  at 2025-07-16 10:30:00
DESCRIPTION:Recent breakthroughs in artificial intelligence, particularly in deep learning, have revolutionized our ability to analyze and interpret biological data, including DNA, RNA, and protein sequences. These advances have led to significant progress in critical fields such as medicine, agriculture, and biotechnology.\nIn this talk, I will present my research on applying deep learning, especially natural language processing (NLP) techniques, to address central challenges in bioinformatics. I will briefly discuss parallels between natural language and biological sequences, along with strategies for adapting NLP methods to account for the key differences between them. I will begin by presenting how the choice of tokenizer can improve model performance while significantly reducing memory usage, a common challenge in deep learning.\nNext, I will discuss our deep learning models for computing multiple sequence alignments, demonstrating that transformer architectures can produce accurate results. I will then present a hybrid model capable of generating ancestral sequences without requiring a specific sequence alignment or phylogenetic tree as input. Our results show that this approach performs competitively and, in some cases, surpasses traditional methods.\nFinally, I will introduce a large language model trained to predict protein function, which outperforms traditional search algorithms, particularly for proteins with low similarity to known sequences. Overall, my research highlights how deep learning and NLP offer powerful new solutions to long-standing challenges in bioinformatics.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 601
UID:eventx6a5a287eecefc10827
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20250717T110000
DTEND;TZID=Asia/Jerusalem:20250717T120000
DTSTAMP;TZID=Asia/Jerusalem:20250717T110000
SUMMARY: MSC  talk by Shelly Golan  Enhancing Consistency-Based Image Generation  at 2025-07-17 11:00:00
DESCRIPTION:The recently introduced Consistency models pose an efficient alternative to diffusion algorithms, enabling rapid and good quality image synthesis. These methods overcome the slowness of diffusion models by directly mapping noise to data, while maintaining a (relatively) simpler training. Consistency models enable a fast one- or few-step generation, but they typically fall somewhat short in sample quality when compared to their diffusion origins. In this work we propose a novel and highly effective technique for post-processing Consistency-based generated images, enhancing their perceptual quality. Our approach utilizes a joint classifier-discriminator model, in which both portions are trained adversarially. While the classifier aims to grade an image based on its assignment to a designated class, the discriminator portion of the very same network leverages the softmax values to assess the proximity of the input image to the targeted data manifold, thereby serving as an Energy-based Model. By employing example-specific projected gradient iterations under the guidance of this joint machine, we refine synthesized images and achieve an improved FID scores on the ImageNet 64x64 dataset for both Consistency-Training and Consistency-Distillation techniques.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom
UID:eventx6a5a287eecf0c10830
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20250717T180000
DTEND;TZID=Asia/Jerusalem:20250717T200000
DTSTAMP;TZID=Asia/Jerusalem:20250717T180000
SUMMARY: CSpecial Event  Join the Faculty's Interested Day - 17.7.25!  at 2025-07-17 18:00:00
DESCRIPTION:Technion Faculty of Computer Science Interested Day\n17.7 | 18:00 | ZOOM Meeting\nThe world is changing at a pace that requires someone to lead the change. At the Technion Faculty of Computer Science, we are nurturing technology leaders, researchers, and entrepreneurs.\nOn the program:\nOn AI and the future world of employmentProf. Danny Raz, Dean of the Faculty\nPanel on &lsquo;Computer Science Studies&rsquo; at the TechnionParticipating: Dean of the Faculty, students, and alumni\n\nWhy Computer Science at the Technion?The Technion Faculty of Computer Science offers more than top-notch studies; it combines academic excellence with direct industry connections, vibrant campus life, and a cohesive community. Students are integrated into projects with leading companies, enjoy events and activities such as music evenings, gaming, competitive teams, and inspiring communities - all in an environment that encourages innovation and creativity.\n\nDevelop AI in an experiential environmentThe Faculty of Computer Science at the Technion combines advanced studies in AI with an unparalleled campus experience.&nbsp;From the classroom to the learning spaces - with a vibrant research environment, social events, band nights, and friends who will become partners on the journey even after graduation.\n\nClimb to the top in studies and beyondStudents at the faculty are not satisfied with perfect code, they strive for excellence beyond the classroom: in academia, in sports, and in any challenge that requires discipline, curiosity, and a drive to lead.\nRegister for the day at the link
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom
UID:eventx6a5a287eecf1d10829
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20250729T113000
DTEND;TZID=Asia/Jerusalem:20250729T123000
DTSTAMP;TZID=Asia/Jerusalem:20250729T113000
SUMMARY: PHD  talk by Tomer Lange  Predictive Garbage Collection in SSDs - or: How Would You Group Your Intervals?  at 2025-07-29 11:30:00
DESCRIPTION:Non-volatile memories (NVMs) have become increasingly common in modern storage systems, offering the advantage of data persistence during power outages. However, several NVM technologies, such as flash and phase-change memory, support only a limited number of data writes, after which memory cells begin to wear out and can no longer reliably store data. Although these limitations have been extensively studied in the systems community, they remain surprisingly unexplored in theoretical models, revealing an unusually wide gap between theory and practice.\nTo bridge this gap, we define and study several optimization problems that capture the fundamental constraints of real-world memory systems. Specifically, this talk focuses on the unique algorithmic challenges arising in the context of SSD management. Building on an interval-based view of the input, we introduce two complementary approaches for optimizing SSD performance. We then demonstrate how these approaches can be applied in a learning-augmented online setting, where only partial (and possibly erroneous) information about the future is available.\nFor this setting, we present two algorithms with tight worst-case guarantees and demonstrate their practical effectiveness through empirical evaluation. Together, our results provide both theoretical and practical insights, and offer new perspectives at the intersection of memory management and online algorithms.\n&nbsp;
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 601
UID:eventx6a5a287eecf3510832
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20250730T113000
DTEND;TZID=Asia/Jerusalem:20250730T123000
DTSTAMP;TZID=Asia/Jerusalem:20250730T113000
SUMMARY: MSC  talk by Shalev Kuba  Online Deduplicated Data Migration  at 2025-07-30 11:30:00
DESCRIPTION:In storage systems, the data migration process periodically remaps files between volumes with the goal of preserving the system&rsquo;s load balance and deduplication efficiency. Previous studies focused on offline selection of files to migrate, a task complicated by the inter-file dependencies introduced by deduplication. However, they did not address the possibility of files entering and leaving the system due to user actions, nor the order between individual file transfers. Our motivational study reveals that na&iuml;ve ordering may create traffic spikes and leave the system in poorly balanced intermediate states. To address these challenges, we present Slide---a novel online migration approach based on sliding windows. Slide takes advantage of long-term planning to maximize deduplication efficiency while maintaining short-term load balance and adapting to system changes. It achieves superior load balancing than alternative approaches while incurring minimal increase in the overall system size.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 301 &amp; Zoom
UID:eventx6a5a287eecf4810831
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20250820T113000
DTEND;TZID=Asia/Jerusalem:20250820T123000
DTSTAMP;TZID=Asia/Jerusalem:20250820T113000
SUMMARY: MSC  talk by Yaniv Wolf  A Data-Driven Approach for Multi-View Surface Reconstruction  at 2025-08-20 11:30:00
DESCRIPTION:Multi-view surface reconstruction, the task of recovering accurate surfaces from multi-view images, has shifted in recent years from classical shape-from-X pipelines that rely on handcrafted feature matching, to fast, data-driven stereo/multi-view stereo correspondence models and emerging 3D foundation models. In parallel, implicit 3D representations such as Neural Radiance Fields and Gaussian Splatting (GS) have revolutionized novel view synthesis, with GS in particular achieving incredible speed and rendering quality. However, extracting reliable geometry from an appearance-only optimized GS representation is a challenging task. Prior works attempt to inject heuristic and/or data-driven geometric priors during the GS optimization phase, often resulting in a tradeoff between rendering quality and geometric accuracy.\nWe propose to avoid this tradeoff, and introduce GS2Mesh, a novel method for incorporating data-driven priors into GS, in a manner that not only avoids damaging the rendering quality, but actually uses the rendering quality to improve the geometric quality. We compose a novel pipeline, in which a fully optimized GS representation is used to manipulate the original multi-view monocular input, into a multi-view consistent stereoscopic input, from which a pre-trained data-driven stereo model can extract accurate geometry. Our pipeline is model-agnostic, and can naturally improve as newer GS and stereo models emerge.\nWe show how GS and stereo work well with each other, and demonstrate our method&rsquo;s state-of-the-art performance both in speed and accuracy on popular 3D reconstruction benchmarks, as well as on in-the-wild videos taken by a standard smartphone camera.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 401 &amp; Zoom
UID:eventx6a5a287eecf5c10834
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20250826T160000
DTEND;TZID=Asia/Jerusalem:20250826T170000
DTSTAMP;TZID=Asia/Jerusalem:20250826T160000
SUMMARY: MSC  talk by Kirill Kutsenok  Sample-Based PCPs  at 2025-08-26 16:00:00
DESCRIPTION:Probabilistically Checkable Proofs (PCPs) allow a verifier to check the validity of a proof by reading only a few symbols chosen at random. While the symbols are chosen at random, classical PCP constructions rely on complex correlations between the different queries. Motivated by applications to proofs over noisy channels, in this work we initiate the study of "sample-based PCPs" in which the verifier's queries are chosen uniformly at random without such correlations.&nbsp;Our main results are constructions of sample-based PCPs:1. &nbsp;For proving the satisfiability of a circuit C of size S, we construct a sample-based PCP of length S*polylog(S) over a constant size alphabet in which the verifier reads \sqrt{S}*polylog(S) random symbols.2. We construct a "zero-knowledge" sample-based PCP with similar parameters to our first result, but for which the sample-based verifier learns essentially nothing beyond the fact that C is satisfiable.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 8 & Zoom
UID:eventx6a5a287eecf7210836
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20250827T133000
DTEND;TZID=Asia/Jerusalem:20250827T143000
DTSTAMP;TZID=Asia/Jerusalem:20250827T133000
SUMMARY: MSC  talk by Hen Davidov  Conformalized survival analysis for LLM safety and clinical decision making  at 2025-08-27 13:30:00
DESCRIPTION:This talk presents methods for uncertainty quantification under partial information, aiming to enable reliable predictions in AI safety and healthcare. At first glance, missing information might seem to make trustworthy predictions impossible. However, I will demonstrate that with appropriate statistical tools, we can achieve reliable predictions even under these challenging conditions.\nI will begin by examining safety evaluation for large language models (LLMs). Specifically, we tackle the problem of estimating time-to-unsafe-sampling&mdash;how many generations are required before a model produces an unsafe response, such as toxic content.The core challenge is that unsafe outputs are extremely rare. For many prompts, no unsafe response appears within any feasible number of samples, making direct estimation impractical. This scarcity creates significant obstacles for reliable safety assessment.Our solution reframes this as a survival analysis problem. We develop a framework that constructs lower bounds on time-to-unsafe-sampling of a given prompt with finite-sample guarantees. By combining conformal prediction with an adaptive, per-prompt sampling strategy, our method delivers formal uncertainty guarantees while efficiently using limited samples. The result is a statistically principled, fine-grained approach to measuring safety risks in generative models.\nI will also discuss how conformal prediction can utilize censored data in clinical trials. Real-world medical datasets frequently provide only partial observations&mdash;we might know either when a patient experienced an outcome or when they left the study, but rarely both.I introduce a new conformal prediction framework designed for this general form of censoring. This approach enables valid and informative lower bounds on survival time, providing reliable patient outcome estimates even when datasets have incomplete follow-up information.\nIn short: partial observations need not prevent trustworthy predictions. Through principled uncertainty quantification, we can transform incomplete data into reliable foresight across critical domains.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Room 401 & Zoom
UID:eventx6a5a287eecf8210839
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20250902T150000
DTEND;TZID=Asia/Jerusalem:20250902T160000
DTSTAMP;TZID=Asia/Jerusalem:20250902T150000
SUMMARY: MSC  talk by Alon Yerushalmi  Perfectly Correct Additive Randomized Encodings  at 2025-09-02 15:00:00
DESCRIPTION:An Additive Randomized Encoding (ARE) for a distributed function f(x1,...,xn) reduces the task of securely computing f to computing the sum of locally encoded inputs.Previous works construct AREs for simple functions such as OR with perfect security but imperfect correctness (namely, the decoder has an incorrect output with small probability), and AREs for general functions with both imperfect correctness and security.\nThis leaves open the existence of ARE for nontrivial functions with perfect correctness, i.e. where the sum of encodings is always decoded correctly to the function's output. We made progress on these questions, obtaining negative results.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 8 &amp; Zoom
UID:eventx6a5a287eecf9610838
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20250903T130000
DTEND;TZID=Asia/Jerusalem:20250903T140000
DTSTAMP;TZID=Asia/Jerusalem:20250903T130000
SUMMARY: ceClub  talk by Roi Bar-Zur (Technion)  CE-Club - Blockchain Analysis with Reinforcement Learning  at 2025-09-03 13:00:00
DESCRIPTION:Blockchains secure trillions of dollars in value while operating in adversarial environments where rational actors pursue economic gain. When economic incentives fail to align with system security, critical vulnerabilities emerge. This talk examines how to identify and defend against these vulnerabilities across different decentralized systems.\nFirst, I will present analysis of optimal selfish mining strategies, where miners manipulate the protocol for profit, particularly in scenarios where previously unexplored attack vectors are created by complex revenue structures and potential collaboration with other miners. I demonstrate how rational participants that deliberately assist attackers for their own gain risk the whole system's security. Building on this analysis, I introduce a novel protocol that, compared to Bitcoin's current implementation, provides stronger resistance to these collaborative selfish mining attacks.\nSecond, I will address vulnerabilities in restaking networks, an emerging paradigm where operators use their locked capital across multiple decentralized applications simultaneously. While restaking improves capital efficiency, the stake reuse introduces new risks. I present a more robust restaking architecture that better balances the economic benefits and potential risks.\nTogether, these contributions advance our understanding of how to align incentives with security in decentralized systems and provide practical solutions for their design.\nPhD Seminar. Supervisor: Prof. Ittay Eyal.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Mayer 1061 & Zoom
UID:eventx6a5a287eecfa710841
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20250910T092900
DTEND;TZID=Asia/Jerusalem:20250910T102900
DTSTAMP;TZID=Asia/Jerusalem:20250910T092900
SUMMARY: MSC  talk by Yaakov Sherma  From Agent Centric to Obstacle Centric Planning: A Makespan-Optimal Algorithm for the Multi-Agent Warehouse Rearrangement Problem  at 2025-09-10 09:29:00
DESCRIPTION:The Multi-Agent Warehouse Rearrangement (MAWR) problem calls for computing agents plans such that they collectively rearrange a warehouse environment from a given layout which species the location of every movable obstacle in the environment, to a goal layout. It is a natural variant of the well-studied Multi-Agent Path Finding (MAPF) and Multi-Agent Pickup and Delivery (MAPD) problems, which have numerous applications in warehouse automation. Similar to MAPF and MAPD, the problem is computationally hard and existing methods forgo any optimality guarantees while computing solutions to the problem. In contrast, in this work we present the first fully-coupled search-based makespan-optimal approach for solving the MAWR problem. This is done by shifting the algorithmic viewpoint from being agent centric to obstacle centric: Common MAPF and MAPD algorithms &nbsp;are agent-centric wherein tasks are assigned to agents, and the agents' paths are planned to fulfill these tasks. In contrast, the approach we present here is obstacle-centric: Our algorithm iterates between two phases: Ph1 planning optimal paths for the movable obstacles using an adaptation of the celebrated CBS MAPF algorithm and Ph2 attempting to realize the movable obstacles paths using a network-flow algorithm. We prove that our approach guarantees minimal-makespan solutions and empirically demonstrate that it achieves better plans than state-of-the-art approaches for MAWR.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom
UID:eventx6a5a287eecfb910846
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20250910T120000
DTEND;TZID=Asia/Jerusalem:20250910T130000
DTSTAMP;TZID=Asia/Jerusalem:20250910T120000
SUMMARY: MSC  talk by Chen Pery  UNet as a Poisson Solver for Photometric Stereo  at 2025-09-10 12:00:00
DESCRIPTION:Photometric Stereo recovers the 3D shape of a surface from multiple images under different lighting directions. The reconstruction is formulated as a variational problem that reduces to solving the Poisson equation, which can be addressed with classical methods such as Jacobi, Gauss&ndash;Seidel, SOR, and multigrid. Simple methods reduce high-frequency errors but struggle with low-frequency ones, whereas multigrid handles both efficiently. I present a UNet&ndash;based model trained only on synthetic data as a fast Poisson solver for face surface reconstruction, achieving accuracy comparable to multigrid while being significantly faster on real face datasets.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom
UID:eventx6a5a287eecfc910847
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20250914T110000
DTEND;TZID=Asia/Jerusalem:20250914T120000
DTSTAMP;TZID=Asia/Jerusalem:20250914T110000
SUMMARY: MSC  talk by Besart Dollma  Coding for Ordered Composite DNA Sequences  at 2025-09-14 11:00:00
DESCRIPTION:To increase the information capacity of DNA storage, composite DNA letters were introduced. We propose a novel channel model for composite DNA in which composite sequences are decomposed into ordered non-composite sequences. The model is designed to handle any alphabet size and composite resolution parameter. We study the problem of reconstructing composite sequences of arbitrary resolution over the binary alphabet under substitution errors. We define two families of error-correcting codes and provide lower and upper bounds on their cardinality. In addition, we analyze the case in which a single deletion error occurs in the channel and present a systematic code construction for this setting. Finally, we briefly discuss the channel's capacity, which remains an open problem.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 601 & Zoom
UID:eventx6a5a287eecfd610833
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20250915T120000
DTEND;TZID=Asia/Jerusalem:20250915T130000
DTSTAMP;TZID=Asia/Jerusalem:20250915T120000
SUMMARY: MSC  talk by Ehud Gordon  Decomposing CLIP's Embedding Space: Towards Improved Interpretability and Control  at 2025-09-15 12:00:00
DESCRIPTION:Vision-Language Models (VLMs) like CLIP have transformed the field by enabling joint reasoning across modalities, zero-shot transfer, and enhanced multimodal alignment. Despite their success and widespread adoption, embeddings derived from CLIP exhibit limitations, including challenges in object binding, relation comprehension, and interpretability due to difficulty in interpretability stemming from entangled feature representations. This work, under Supervision of Prof. Guy Gilboa, investigates methods for decomposing and analyzing CLIP's embedding space, employing various statistical and decomposition techniques. Our approach seeks to enhance performance, interpretability, and robustness across multiple applications, including image classification and editing. By addressing foundational representational challenges, this research contributes towards a deeper understanding of multimodal embedding geometry and advances the interpretability of modern VLMs.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Meyer Building 1061 &amp; Zoom
UID:eventx6a5a287eecfe510848
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20250915T143000
DTEND;TZID=Asia/Jerusalem:20250915T153000
DTSTAMP;TZID=Asia/Jerusalem:20250915T143000
SUMMARY: MSC  talk by Luai Taha  Chemical Vapor Deposition (CVD) Growth of Monolayer MoS2: Impact of Growth Promoters on Morphology and Electrical Performance  at 2025-09-15 14:30:00
DESCRIPTION:Two-dimensional (2D) materials, such as transition metal dichalcogenides (TMDs), have attracted significant attention for their potential in next-generation electronic and optoelectronic devices due to their atomically thin structure and unique electrical properties. Among them, monolayer MoS2 stands out as a promising semiconductor for low-power field-effect transistors (FETs).\nIn this work, we investigate the growth of monolayer MoS2 using chemical vapor deposition (CVD) with MoO3 and sulfur precursors. The effects of growth promoters, including NaCl and glass substrates, are systematically studied, revealing their influence on domain size, film continuity, and monolayer coverage.\nA reliable transfer method is implemented to integrate the MoS2 films onto device compatible substrates. Comprehensive characterization confirms high structural and optical quality, and FETs fabricated from the transferred MoS2 exhibit promising electrical performance. These results provide valuable insights into the scalable growth&ndash;performance relationship and support the integration of CVD-grown MoS2 into future low-power electronic platforms.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Mayer 1061
UID:eventx6a5a287eecff410842
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20250917T090000
DTEND;TZID=Asia/Jerusalem:20250917T100000
DTSTAMP;TZID=Asia/Jerusalem:20250917T090000
SUMMARY: MSC  talk by Yael Shabtay  Faster Partially-Dynamic Size Estimation  at 2025-09-17 09:00:00
DESCRIPTION:Dynamic graphs capture systems whose connectivity evolves. As edges are inserted and deleted, recomputing answers after each change becomes impractical, motivating algorithms that maintain information as the graph evolves. We study two counting problems on dynamic directed graphs, considered for every vertex simultaneously: (i) the size of its reachable set and (ii) the number of vertices within distance at most d. We present a simple randomized method that maintains (1&plusmn;&epsilon;) estimates with polylogarithmic update and query times in the incremental model, with analogous guarantees for the decremental model on DAGs. Complementing these algorithms, we prove a conditional limitation, which motivates approximation: under SETH, no partially dynamic algorithm can simultaneously achieve sublinear update and query time for maintaining the exact sizes of reachable sets, even on DAGs and even when the entire update sequence is known in advance.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom
UID:eventx6a5a287eed00310844
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20250917T113000
DTEND;TZID=Asia/Jerusalem:20250917T123000
DTSTAMP;TZID=Asia/Jerusalem:20250917T113000
SUMMARY: MSC  talk by Saar Tzour-Shaday  Mini-Batch Robustness Verification of Deep Neural Networks  at 2025-09-17 11:30:00
DESCRIPTION:Neural network image classifiers are ubiquitous in many safety-critical applications. However, they are susceptible to adversarial attacks. To understand their robustness to attacks, many local robustness verifiers have been proposed to analyze 𝜖-balls of inputs. Yet, existing verifiers introduce a long analysis time or lose too much precision, making them less effective for a large set of inputs. In this work, we propose a new approach to local robustness: group local robustness verification. The key idea is to leverage the similarity of the network computations of certain 𝜖-balls to reduce the overall analysis time. We propose BaVerLy, a sound and complete verifier that boosts the local robustness verification of a set of 𝜖-balls by dynamically constructing and verifying mini-batches. BaVerLy adaptively identifies successful mini-batch sizes, accordingly constructs mini-batches of 𝜖-balls that have similar network computations, and verifies them jointly. If a mini-batch is verified, all 𝜖-balls are proven robust. Otherwise, one 𝜖-ball is suspected as not being robust, guiding the refinement. In the latter case, BaVerLy leverages the analysis results to expedite the analysis of that 𝜖-ball as well as the other 𝜖-balls in the batch. We evaluate BaVerLy on fully connected and convolutional networks for MNIST and CIFAR-10. Results show that BaVerLy scales the common one by one verification by 2.3x on average and up to 4.1x, in which case it reduces the total analysis time from 24 hours to 6 hours.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Mayer 1061, Zoom
UID:eventx6a5a287eed06a10837
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20250917T143000
DTEND;TZID=Asia/Jerusalem:20250917T153000
DTSTAMP;TZID=Asia/Jerusalem:20250917T143000
SUMMARY: MSC  talk by Barak Biber  Ruse - Concrete Graphs Synthesis  at 2025-09-17 14:30:00
DESCRIPTION:In the world of Program Synthesis, bottom-up enumeration with observational equivalence is a common approach&nbsp;for generating programs.&nbsp;However, classical observational equivalence has a strong assumption that all generated programs are pure,&nbsp;a constraint that is particularly limiting for object-oriented languages.\nWe introduce Concrete Graph Logic, an extension of Hoare Logic, to address this limitation.&nbsp;In this logic, we represent the program context and its objects using Object Graphs&nbsp;and introduce a new rule called the embedding rule, which replaces the consequence rule in Hoare Logic.\nThis rule defines when an embedding of pre-condition and post-condition graphs in a larger graph&nbsp;maintains the validity of the Hoare triplet.&nbsp;Using this rule and the standard composition rule, we can compose two Hoare triplets with compatible conditions.\nWe also present an algorithm that, given two Hoare triplets, finds a valid embedding that allows&nbsp;for their composition, if one exists.&nbsp;This algorithm is implemented in Ruse, a new program synthesizer capable of generating&nbsp;mutating object-oriented JavaScript programs.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 601 & Zoom
UID:eventx6a5a287eed07f10843
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20250929T110000
DTEND;TZID=Asia/Jerusalem:20250929T120000
DTSTAMP;TZID=Asia/Jerusalem:20250929T110000
SUMMARY: MSC  talk by Ido Sobol  Enhancing Zero-Shot Novel View Synthesis via Attention Map Filtering  at 2025-09-29 11:00:00
DESCRIPTION:Generating realistic images from arbitrary views based on a single source image remains a significant challenge in computer vision, with broad applications ranging from e-commerce to immersive virtual experiences. Recent advancements in diffusion models, particularly the Zero-1-to-3 model, have been widely adopted for generating plausible views, videos, and 3D models. However, these models still struggle with inconsistencies and implausibility in new views generation, especially for challenging changes in viewpoint. We propose Zero-to-Hero, a novel test-time approach that enhances view synthesis by manipulating attention maps during the denoising process of Zero-1-to-3. By drawing an analogy between the denoising process and stochastic gradient descent (SGD), we implement a filtering mechanism that aggregates attention maps, enhancing generation reliability and authenticity. This process improves geometric consistency without requiring retraining or significant computational resources. Additionally, we modify the self-attention mechanism to integrate information from the source view, reducing shape distortions. These processes are further supported by a specialized sampling schedule. Additionally, we demonstrate the general applicability and effectiveness of Zero-to-Hero in multi-view, and image generation conditioned on semantic maps and pose.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 601 &amp; Zoom &nbsp;
UID:eventx6a5a287eed09010849
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20251016T123000
DTEND;TZID=Asia/Jerusalem:20251016T133000
DTSTAMP;TZID=Asia/Jerusalem:20251016T123000
SUMMARY: MSC  talk by Gilad Altshuler  Beyond Linear Summation: Three-Body RNNs For Modeling Complex Neural And Biological Systems  at 2025-10-16 12:30:00
DESCRIPTION:Recurrent neural networks (RNNs) are a widely used models in neuroscience, whose architecture is inspired by the abundance of lateral connections in cortex. The basic computation in RNNs is a linear summation of inputs, followed by a pointwise nonlinearity. Outside neuroscience, similar models are used to describe ecological, transcriptional, and other biological networks. Real biological networks, however, exhibit many cooperative effects such as dendritic gating, neuromodulation, glia&ndash;neuron coupling and transcription factor dimerization. We ask how recurrent dynamics change when explicit three-body (quadratic) interactions are included. We introduce the Three-Body Recurrent Neural Network (TBRNN) and obtain four main results. First, we prove that TBRNNs are universal approximators for open dynamical systems. Second, we extend low-rank RNN theory to this setting, both to infer connectivity and to design functionality. Third, training on canonical neuroscience tasks reveals solution families that differ in their geometry from those found by standard RNNs, indicating that higher-order interactions reorganize the accessible dynamical regimes rather than merely re-parameterizing pairwise models. Fourth, we develop a practical model-comparison procedure that detects three-body signatures directly from trajectories, providing a way to infer when higher-order structure is present in recorded systems. Altogether, our work expands the space of dynamical models in neuroscience while also allowing other fields of biology to benefit from insights developed in neuroscience.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom
UID:eventx6a5a287eed1ba10851
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20251021T093000
DTEND;TZID=Asia/Jerusalem:20251021T103000
DTSTAMP;TZID=Asia/Jerusalem:20251021T093000
SUMMARY: MSC  talk by Ofir Feder  Automatic Detection of Structural Hotspots in the Calcium Channel Protein  at 2025-10-21 09:30:00
DESCRIPTION:The function of a protein is largely determined by its ability to undergo structural changes and shift between various transitional states. These conformational transitions are often induced when proteins interact with their environment, and they can also be induced by mutations to key amino acids (residues) in the protein. In most cases, the majority of changes between transition states can be attributed to a small number of regions in the protein, called "structural hotspots.'' Identifying and studying the changes in the structural hotspots is a key task in understanding the connection between a protein's structure and function. To study the structure of a protein, biologists often rely on 3D structural data obtained from protein crystallography or cryo-EM studies. Such studies are, however, highly tedious, expensive, and time-consuming. Moreover, they are rarely able to capture the different transition states of the studied protein without prior knowledge of the specific stabilizing mutations of each conformation. Additionally, manual inspection of a large number of 3D structures is tedious and error-prone and may be prohibitively time-consuming. In this work, I address this problem in-silico.&nbsp;\nI formally define the problem of hotspot detection, I propose localmatch, an algorithm for detecting a protein's hotspots given a set of 3D structures, and I study the application of this algorithm to the Orai1 calcium channel. To this end, I perform a literature review to collect crystallographic data of different conformations, along with labeled hotspots of the Orai1 protein. I validate my approach by automatically detecting all known hotspots from the literature when running my algorithm on existing crystallographic data in the top 3% of amino acids, significantly outperforming alternative baselines. Furthermore, I demonstrate localmatch's ability to identify the hotspots from the literature based on 3D structures of mutations predicted by AlphaFold, where it detects four of the five hotspots within the top 4% top amino acids.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 201
UID:eventx6a5a287eed1dc10853
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20251021T110000
DTEND;TZID=Asia/Jerusalem:20251021T120000
DTSTAMP;TZID=Asia/Jerusalem:20251021T110000
SUMMARY: PHD  talk by Shunit Agmon  Computational Tools for Mitigating Human Biases in Data Collection and Analysis  at 2025-10-21 11:00:00
DESCRIPTION:Human biases can influence multiple stages of data management, from collection to interpretation and use. This thesis aims to develop and investigate techniques for mitigating such biases using computational methods. We begin by describing approaches to mitigate the effects of bias in the data collection phase, using tools from natural language processing. We then proceed to develop tools for assessing the robustness of data-driven claims.\nIn many cases, the data itself encodes human biases, which in turn propagate into the machine learning models trained on it. A prominent example arises in clinical trials, where women have been historically underrepresented. This bias harms the performance of language models and downstream predictions. Rather than erasing gender information, we propose methods that adjust embeddings to reflect underrepresentation while preserving medically meaningful distinctions, thus increasing prediction accuracy for women patients without degrading the accuracy for men patients.\nEven when the underlying data is not inherently biased, it can still be selectively interpreted to support predetermined conclusions - a phenomenon known as cherry-picking. To inspect the robustness of claims based on a database, we propose two complementary methods. The first is a schema-based approach: given a claim and a database, we identify natural views or subpopulations of the database where the claim holds. The naturalness and the number of supporting views for the claim and its opposite can help the user assess the correctness of the claim. The second method is data-driven: we quantify and explain the deviation of a dataset from an expected monotonic trend (such as salaries increasing with education) by finding a minimal repair: the smallest possible set of tuples whose removal restores the expected trend.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 201 &amp; Zoom
UID:eventx6a5a287eed1f110852
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20251028T160000
DTEND;TZID=Asia/Jerusalem:20251028T170000
DTSTAMP;TZID=Asia/Jerusalem:20251028T160000
SUMMARY: ceClub  talk by Barak Gerstein (Technion)   CE-Club - Making Congestion Control Robust to Per-Packet Load Balancing in Datacenters  at 2025-10-28 16:00:00
DESCRIPTION:Per-packet load-balancing approaches are increasingly deployed in datacenter networks. However, their combination with existing congestion control algorithms (CCAs) may lead to poor performance and even throughput collapse.\nIn this talk, I first model the throughput collapse of a wide array of CCAs when some of the paths are congested. I explain that since CCAs are typically designed for single-path routing, their estimation function focuses on the latest feedback and mishandles feedback that reflects multiple paths. I propose to use a median feedback that is more robust to the varying signals that come with multiple paths. I introduce MSwift, which applies this principle to make Google&rsquo;s Swift CCA robust to multi-path routing while keeping its incast tolerance and single-path performance. Finally, I demonstrate that MSwift improves the 99th-percentile FCT by up to 25%, both with random packet spraying and adaptive routing.\nM.Sc. student under the supervision of Prof. Isaac Keslassy and Prof. Mark Silberstein.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Mayer 1061 &amp; Zoom
UID:eventx6a5a287eed20410856
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20251104T123000
DTEND;TZID=Asia/Jerusalem:20251104T143000
DTSTAMP;TZID=Asia/Jerusalem:20251104T123000
SUMMARY: pixel-club  talk by Noam Elata   Pixel Club - Diffusion Models for Posterior Sampling and Adaptive Sensing  at 2025-11-04 12:30:00
DESCRIPTION:Diffusion models have emerged as the leading approach for high-quality image synthesis and demonstrate exceptional versatility in solving inverse problems through their powerful learned image priors. In this seminar, we explore how these generative priors enable adaptive compressed sensing for real-world active acquisition applications, including MRI and CT imaging, where intelligent measurement selection can dramatically reduce scan times while preserving reconstruction quality.\nWe further demonstrate how these same principles extend naturally to image compression, leveraging the diffusion prior to achieve efficient encoding and high-fidelity reconstruction.\nMotivated by limitations in existing posterior sampling methods, we introduce a novel model architecture specifically designed for inverse problems that is both theoretically justified and computationally efficient. Collectively, these contributions establish a unified framework for deploying diffusion models across medical imaging, image compression, and image restoration, advancing both the practical applicability and theoretical foundations of generative models for inverse problems.\nNoam Elata is a Ph.D. candidate under the supervision of Prof. Michael Elad and Prof. Tomer Michaeli.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:506, Zisapel Building &amp; Zoom
UID:eventx6a5a287eed21910855
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20251104T143000
DTEND;TZID=Asia/Jerusalem:20251104T163000
DTSTAMP;TZID=Asia/Jerusalem:20251104T143000
SUMMARY: colloq  talk by Steven M. LaValle, Professor of Robotics and Virtual Reality (University of Oulu & University of Illinois)  Fundamental Challenges in Robotics and Embodied AI  at 2025-11-04 14:30:00
DESCRIPTION:The field of robotics is wildly exciting and rapidly gaining worldwide attention, yet it is often an enigma in terms of its scope and scientific foundations. &nbsp;Robotics involves the design, programming, and analysis of movable machines that accomplish useful work through sensing and manipulation of the surrounding world. &nbsp;Throughout the decades it has been varyingly viewed as an application field of more mature disciplines such as computer science (AI, algorithms, machine learning) and mechanical engineering (kinematics, dynamics, nonlinear control). This talk will argue that robotics has its own unique and growing scientific core, with deep questions and modelling challenges that should inspire new directions in computer science, engineering, and even pure mathematics.\nWe will start with a Turing-inspired way to view robotics or embodied AI, leading to some of our recent results that characterize minimally sufficient amounts of sensing, actuation, or computation that are required to solve physical tasks. Questions addressed include: How are learning, planning, and control related? How do we know when it is impossible to solve a task? &nbsp;When will learning fail, even with an infinite amount of data? &nbsp;Does a universal action sequence exist that would cause a robot to solve any possible task without modification? &nbsp;How important are semantics and representations? &nbsp;Interspersed throughout the talk will be results and perspective from my research in the field over three decades, produced with many inspiring students, mentors, and collaborators.\nBio: Steven M. LaValle has been Professor of Computer Science and Engineering, in Robotics and Virtual Reality, at the University of Oulu, Finland since 2018. &nbsp;Since 2001, he has been a professor in the Department of Computer Science at the University of Illinois. &nbsp;He has also held positions at Stanford University and Iowa State University. &nbsp;His research interests include robotics, virtual reality, sensor fusion, planning algorithms, computational geometry, and control theory. &nbsp;In research, he is mostly known for his introduction of the Rapidly exploring Random Tree (RRT) algorithm, which is widely used in robotics and other engineering fields. He also authored the books Planning Algorithms, Sensing and Filtering, and Virtual Reality. &nbsp;He currently leads an Advanced Grant project from the European Research Council on the Foundations of Perception Engineering. &nbsp;\nWith regard to industry, he was an early founder and chief scientist of Oculus VR, acquired by Facebook for $3 billion in 2014, where he developed patented tracking technology for consumer virtual reality and led a team of perceptual psychologists to provide principled\napproaches to virtual reality system calibration, health and safety, and the design of comfortable user experiences. &nbsp;From 2016 to 2017, he was a Vice President and Chief Scientist of VR/AR/MR at Huawei Technologies, where he was a leader in mobile product development on a global scale. &nbsp;He has worked as an angel investor and adviser to startups in robotics and virtual reality.\nTechnion Host: Oren Salzman
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 337
UID:eventx6a5a287eed22c10854
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20251105T130000
DTEND;TZID=Asia/Jerusalem:20251105T140000
DTSTAMP;TZID=Asia/Jerusalem:20251105T130000
SUMMARY: Theory Semina  talk by Bogdan Chornomaz (Technion)  Theory Seminar: Topological tools in PAC learning  at 2025-11-05 13:00:00
DESCRIPTION:In the past few years, together with several collaborators, we have developed a framework linking PAC learning theory to topological combinatorics. At its core lies the notion of the spherical dimension of a concept class. Since the most established complexity measure in PAC learning is VC dimension, a natural question arises: can spherical dimension be bounded in terms of VC dimension? This question is compelling from both the learning-theoretic and the topological standpoint. In this talk, I will outline how the connection between these two dimensions arises, and survey what we know, and don't know, about their relation.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 401
UID:eventx6a5a287eed24710858
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20251105T173000
DTEND;TZID=Asia/Jerusalem:20251105T192900
DTSTAMP;TZID=Asia/Jerusalem:20251105T173000
SUMMARY: CSpecial Event  Opening Of The Academic Year For Undergraduate Students  at 2025-11-05 17:30:00
DESCRIPTION:Opening of the academic year for undergraduate students\nWednesday, November 5, 2025, Taub Building, Room 337\n17:30 &ndash; Guide for the Beginning Student for First-Year Students18:30 &ndash; All Undergraduate Students Join Mingling. Food and Drink on Us\nTo Register: https://forms.gle/bFTBuyT1H89QKNgd6\nLet's Start the Year in a Good Atmosphere!
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Room 337
UID:eventx6a5a287eed25a10862
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20251110T100000
DTEND;TZID=Asia/Jerusalem:20251110T110000
DTSTAMP;TZID=Asia/Jerusalem:20251110T100000
SUMMARY: PHD  talk by Itshak Blau  Leveraging Task Structure for Robustness and Generalization in Modern Machine Learning  at 2025-11-10 10:00:00
DESCRIPTION:Modern machine learning models often struggle to remain reliable under challenging conditions such as distribution shifts, adversarial perturbations, or limited data. My research focuses on improving robustness and generalization by leveraging task-specific structure at inference time, without requiring additional training or data. I present methods that adapt either the inputs or the context of pretrained models to better align with the underlying task. In the visual domain, these approaches enhance robustness by transforming or projecting inputs toward meaningful class- or data-manifold representations. In the language domain, they refine prompt representations to extract more effective information from few-shot examples. Together, these contributions demonstrate a unified principle: carefully adapting representations to reflect task-relevant structure can substantially improve the reliability and generalization of modern machine learning systems across both vision and language.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom
UID:eventx6a5a287eed36110857
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20251110T123000
DTEND;TZID=Asia/Jerusalem:20251110T143000
DTSTAMP;TZID=Asia/Jerusalem:20251110T123000
SUMMARY: CSpecial Event  Invitation to the Graduate Studies Open Day - Spring Semester 2025/26  at 2025-11-10 12:30:00
DESCRIPTION:You are invited to an Information Session about the Master&rsquo;s program, intended for outstanding Bachelor&rsquo;s graduates.\nThe session will take place on Monday, November 10, 2025, at 12:3 pm, in Auditorium 012, Floor 0, Taub Building.\nPlease register using the link so we can prepare accordingly.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Auditorium 012, Floor 0, Taub Building
UID:eventx6a5a287eed37c10850
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20251110T183000
DTEND;TZID=Asia/Jerusalem:20251110T203000
DTSTAMP;TZID=Asia/Jerusalem:20251110T183000
SUMMARY: CSpecial Event  Come Be Part Of The Faculty's Capture The Flag - CTF Group!! 10.11.25  at 2025-11-10 18:30:00
DESCRIPTION:Come be part of the faculty's Capture The Flag - CTF group!!The first meeting this year will be held on Monday, November 10th at 6:30 PM at Taub 9.\nCapture The Flag competitions are challenges on cyber and information security topics in a variety of subjects - cryptography, reverse engineering, web, forensics and more.\nThe Technipwn faculty CTF group will hold biweekly meetings to experiment with solving challenges, competitions and fun.\nThe meetings are suitable for beginners and experienced participants, from the faculty and the Technion.&nbsp;Of course, there will be refreshments as is tradition\nEveryone is welcome, even without prior experience. To register: https://docs.google.com/.../1FAIpQLSfMcQo8zgfexK.../viewform\nWe look forward to seeing you!
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 9
UID:eventx6a5a287eed38e10863
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20251111T090000
DTEND;TZID=Asia/Jerusalem:20251111T100000
DTSTAMP;TZID=Asia/Jerusalem:20251111T090000
SUMMARY: PHD  talk by Eden Saig  Machine Learning in the Presence of People: Dynamics and Incentives  at 2025-11-11 09:00:00
DESCRIPTION:As machine learning becomes increasingly central in society, algorithms gain the ability to shape the outcomes they predict. The statistical limitations of prediction models, together with rising costs of training and inference, create scarcity and information-asymmetry conditions that fuel strategic and competitive behaviors.\nThese behaviors violate the assumptions of traditional learning methods, and may produce unintended and possibly harmful outcomes. We study learning algorithms as components embedded in social and economic systems, and use tools from dynamical systems, contract design, and evolutionary game theory to analyze the failure modes that may emerge around them.\nBuilding on this analysis, we develop algorithmic and mechanism-design approaches that reduce exploitation, improve economic efficiency, and steer systems toward socially favorable outcomes. Together, our results demonstrate how an integrative, behavior-aware approach can inform both theory and practice.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom
UID:eventx6a5a287eed3a210861
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20251111T170000
DTEND;TZID=Asia/Jerusalem:20251111T183000
DTSTAMP;TZID=Asia/Jerusalem:20251111T170000
SUMMARY: CSpecial Event  talk by Prof. Eitan Yaakobi  Prof. Eitan Yaakobi's Talk At TEDx Technion  at 2025-11-11 17:00:00
DESCRIPTION:We&rsquo;re proud to share that Prof. Eitan Yaakobi, a member of our faculty, will be speaking at TEDx Technion, alongside other leading Technion professors pushing the boundaries of innovation.\nProf. Yaakobi&rsquo;s talk, &ldquo;The Future of Data Storage is in Our DNA,&rdquo; explores how biology could hold the key to storing the world&rsquo;s growing volumes of data.\nWe invite you to join the online event and support Prof. Eitan Yaakobi and the Technion community as they showcase ideas that challenge the status quo and reshape what&rsquo;s possible.\nJoin us on November 11, 17:00 IST: https://tedxtechnion.com/watch-live&nbsp;
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:tedxtechnion
UID:eventx6a5a287eed3b510866
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20251112T110000
DTEND;TZID=Asia/Jerusalem:20251112T120000
DTSTAMP;TZID=Asia/Jerusalem:20251112T110000
SUMMARY: MSC  talk by Romy Peled  Large Lemma Miners: Can LLMs do Induction Proofs for Hardware?  at 2025-11-12 11:00:00
DESCRIPTION:Abstract English: Large Language Models (LLMs) have shown potential in solving mathematical tasks, as demonstrated by their performance on Math Olympiad problems. Motivated by this, we study whether LLMs can be leveraged to generate proofs by induction for hardware verification.\nHardware verification is a challenging task that requires experienced formal verification (FV) engineers. One of the most useful techniques FV engineers use is proof by induction: they prove auxiliary lemmas such that, when used as assumptions, the property of interest becomes inductive, and hence manageable for automatic formal verification tools.&nbsp;\nIn this work, we show that LLMs can replicate some of this process by suggesting lemmas that help construct a proof by induction for hardware verification, indicating that the mathematical reasoning abilities of LLMs may hold industrial value.&nbsp;\nWe present a neurosymbolic approach that includes two prompting frameworks to generate candidate invariants, which are checked using a formal, symbolic tool. Our results indicate that with sufficient reprompting, LLMs are able to generate inductive arguments for mid-size open-source RTL designs. For 87% of our problem set, at least one of the prompt setups succeeded in producing a provably correct inductive argument.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 601
UID:eventx6a5a287eed3c710860
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20251112T123000
DTEND;TZID=Asia/Jerusalem:20251112T143000
DTSTAMP;TZID=Asia/Jerusalem:20251112T123000
SUMMARY: CSpecial Event  Mobileye Spotlight Day  at 2025-11-12 12:30:00
DESCRIPTION:Mobileye is coming to meet you at the faculty!!Next Wednesday 12.11.25 starting at 12:30 at Taub, lobby level.\nOn the program: a meeting with the recruitment team, the development teams and surprises.\nWaiting for you!
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub Lobby, Floor 0
UID:eventx6a5a287eed3d910859
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20251112T130000
DTEND;TZID=Asia/Jerusalem:20251112T140000
DTSTAMP;TZID=Asia/Jerusalem:20251112T130000
SUMMARY: Theory Semina  talk by Boaz Menuhin  Theory Seminar: Shuffling Cards When You Are of Very Little Brain: Low Memory Generation of Permutations  at 2025-11-12 13:00:00
DESCRIPTION:How can we generate a permutation of the numbers 1 through n so that it is hard to guess the next element given the history so far? The twist is that the generator of the permutation (the &ldquo;Dealer&rdquo;) has limited memory, while the &ldquo;Guesser&rdquo; has unlimited memory. With unbounded memory (actually n bits suffice), the Dealer can generate a truly random permutation where ln n is the expected number of correct guesses.\nWe will show tight bounds for the relationship between the guessing probability and the memory m required to generate the permutation. We suggest a method for an m-bit Dealer that operates in constant time per turn, and any Guesser can pick correctly only O(n/m + log m) cards in expectation. The method is fully transparent, requiring no hidden information from the Dealer (i.e., it is &rdquo;open book&rdquo; or &rdquo;whitebox&rdquo;).\nWe show that this bound is the best possible, even with secret memory. Specifically, for any m-bit Dealer, there is a (computationally powerful) Guesser that achieves &Omega;(n/m+log m) correct guesses in expectation. We point out that the assumption that the Guesser is computationally powerful is necessary: under cryptographic assumptions, there exists a low-memory Dealer that can fool any computationally bounded Guesser.\nJoint work with Moni Naor, to appear in FOCS 2025.https://arxiv.org/abs/2505.01287&nbsp;
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 401
UID:eventx6a5a287eed3e810864
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20251117T183000
DTEND;TZID=Asia/Jerusalem:20251117T203000
DTSTAMP;TZID=Asia/Jerusalem:20251117T183000
SUMMARY: CSpecial Event  Come Be Part Of The Faculty's Capture The Flag - CTF Group!! 17.11.25  at 2025-11-17 18:30:00
DESCRIPTION:Come be part of the faculty's Capture The Flag - CTF group!!The meeting will be held on Monday, November 17th at 6:30 PM at Taub 9.\nCapture The Flag competitions are challenges on cyber and information security topics in a variety of subjects - cryptography, reverse engineering, web, forensics and more.\nThe Technipwn faculty CTF group will hold biweekly meetings to experiment with solving challenges, competitions and fun.\nThe meetings are suitable for beginners and experienced participants, from the faculty and the Technion.&nbsp;Of course, there will be refreshments as is tradition\nEveryone is welcome, even without prior experience. To register: https://forms.gle/bc2RSZcUscJ7daBh9\nWe look forward to seeing you!
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 9
UID:eventx6a5a287eed3fb10869
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20251118T113000
DTEND;TZID=Asia/Jerusalem:20251118T133000
DTSTAMP;TZID=Asia/Jerusalem:20251118T113000
SUMMARY: pixel-club  talk by Neta Shaul (Weizmann Institute of Science)  Pixel Club - Transition Matching: Scalable and Flexible Generative Modeling  at 2025-11-18 11:30:00
DESCRIPTION:Transition matching (TM) replaces the infinitesimal-timestep kernels from Flow Matching/Diffusion with a generative model, advancing both flow/diffusion and autoregressive models. TM variants achieve state-of-the-art text-to-image generation.\nNeta Shaul is a PhD student at the Weizmann Institute of Science under the supervision of Prof. Yaron Lipman. His research focuses on developing and advancing scalable modeling frameworks for generative models. He is interested in a variety of data types from both discrete and continuous domains (text, images, videos, proteins, etc).
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:506, Zisapel Building
UID:eventx6a5a287eed40c10867
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20251119T123000
DTEND;TZID=Asia/Jerusalem:20251119T143000
DTSTAMP;TZID=Asia/Jerusalem:20251119T123000
SUMMARY: CSpecial Event  NVIDIA is Coming to Meet You at the Faculty!! 19.11.25  at 2025-11-19 12:30:00
DESCRIPTION:Wednesday, November 19, between 12:30-14:30 at Taub 2 and lobby taub.\nAn opportunity to hear about new technologies and consult on employment options:13:00 DOCA &ndash; AI for developers\nLecturer: Eyal Grover, Group Manager in the Software Organization, AI and DevOps, NVIDIA Israel, and a panel with the company's engineers.\nTo register for the lecture and the panel, click on the link
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 2 and lobby taub
UID:eventx6a5a287eed41d10873
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20251119T130000
DTEND;TZID=Asia/Jerusalem:20251119T140000
DTSTAMP;TZID=Asia/Jerusalem:20251119T130000
SUMMARY: Theory Semina  talk by Orr Fischer (Bar Ilan University)  Theory Seminar: The Surprising Graph Behind K-Nearest Neighbors and Contrastive Learning  at 2025-11-19 13:00:00
DESCRIPTION:Contrastive learning is a comparative-based learning paradigm in which labelers are given triplet queries (A,B,C) of data points, and answer whether in their opinion point A is "closer" to B than C, or vice versa. E.g. if a labeler is given three pictures (A = Lion, B = Tiger, C = Dog), they would say that a lion is closer to a tiger than to a dog. In this talk, we discuss the dimensionality of satisfying m contrastive learning constraints in an l_p metric. In particular, we answer the following question:&nbsp;What is a sufficiently large dimension d, such that given any m (non-contradictory) constraints of the form "dist(A,B) &lt; dist(A,C)", can we find an embedding into R^d such that all constraints are satisfied? (under the l_p distance). We show that e.g. for p=2 that d = Theta(sqrt(m)) is both necessary and sufficient.&nbsp;Relatedly, we use our techniques to answer the following natural question for the k-nearest neighbor problem:&nbsp;What is a sufficiently large dimension d, such that given any metric D on n points, can we find an embedding into R^d that preserves the k-nearest neighbors of each point of D? (under the l_p distance). We show that d = O~(poly(k)) is always sufficient.&nbsp;Our methods are a combination of combinatorial and algebraic techniques. At the heart of our algorithms is an analysis of a graph which we refer to as the constraint graph, whose properties are associated with the dimension bounds we obtain - and in particular, its arboricity (an important measure of density).
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 401
UID:eventx6a5a287eed43110870
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20251120T130000
DTEND;TZID=Asia/Jerusalem:20251120T140000
DTSTAMP;TZID=Asia/Jerusalem:20251120T130000
SUMMARY: MSC  talk by Ran Levinstein  Forgetting Rates in Continual Learning  at 2025-11-20 13:00:00
DESCRIPTION:We study forgetting rates, the loss of earlier knowledge as new tasks arrive in continual learning. We provide improved bounds under random orderings, optimal rates via regularization, and an analysis of greedy task ordering.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 601 &amp; Zoom
UID:eventx6a5a287eed44610871
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20251123T130000
DTEND;TZID=Asia/Jerusalem:20251123T140000
DTSTAMP;TZID=Asia/Jerusalem:20251123T130000
SUMMARY: MSC  talk by Tom Rahav  Offline Detection of RAG Data Poisoning  at 2025-11-23 13:00:00
DESCRIPTION:Retrieval-Augmented Generation (RAG) systems&nbsp;enhance large language models (LLMs) with external knowledge but are known to be vulnerable to optimization-based&nbsp;data poisoning, where adversaries inject adversarially augmented passages into the knowledge base.\nSuch attacks are designed to bias retrieval and generation, leading the system to produce misleading or adversarially steered responses. Existing defenses rely on query-specific passage filtering during inference at either retrieval or generation, inherently incurring latency and redundantly evaluating passages across queries. This results in substantial runtime overhead, which further increases when additional passages are retrieved as&nbsp;substitutions.\nWe present JUDO, a query-agnostic indexing-time defense that detects and filters poisoned passages at ingestion by measuring the semantic instability that arises when adversarial trigger tokens are removed. We demonstrate that clean passages remain stable under such perturbations, whereas optimized poisoned passages exhibit erratic embedding shifts. Applied once per passage, JUDO filters malicious content without altering the inference pipeline or incurring runtime cost.\nWe compare our approach to previous defenses against known poisoning attacks and across multiple retrieval and QA benchmarks, achieving state-of-the-art results of 0.8&ndash;0.9 F1 and a 70% reduction in attack success rate, thereby providing a practical, zero-overhead defense for securing RAG systems against optimization-based poisoning.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom
UID:eventx6a5a287eed45710868
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20251124T183000
DTEND;TZID=Asia/Jerusalem:20251124T203000
DTSTAMP;TZID=Asia/Jerusalem:20251124T183000
SUMMARY: CSpecial Event  Come Be Part Of The Faculty's Capture The Flag - CTF Group!! 24.11.25  at 2025-11-24 18:30:00
DESCRIPTION:Come be part of the faculty's Capture The Flag - CTF group!!The meeting will be held on Monday, November 24th at 6:30 PM at Taub 9 for beginners and Taub 8 for advanced.\nThe topic of the meeting is finding weaknesses in code programs and exploiting them in order to take control of the program.\nEveryone is welcome, even without previous experience - we look forward to seeing you!\nCommunity link:https://chat.whatsapp.com/CDsSNy44Vzl2AIrdvETqge?mode=wwt&nbsp;
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 8 / Taub 9
UID:eventx6a5a287eed46a10883
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20251125T113000
DTEND;TZID=Asia/Jerusalem:20251125T133000
DTSTAMP;TZID=Asia/Jerusalem:20251125T113000
SUMMARY: pixel-club  talk by Sagie Benaim (Hebrew University of Jerusalem)  Pixel Club - Distilling Foundation Models for a Semantic Reconstruction of the 3D World  at 2025-11-25 11:30:00
DESCRIPTION:Enabling machines to reconstruct and semantically understand the 3D world is a fundamental goal for applications such as robotics, autonomous systems, and immersive telepresence. While recent advances achieve photorealistic 3D reconstruction and novel-view synthesis, this talk presents a framework that goes beyond visual accuracy to build scenes that are semantically rich. By distilling the vast knowledge of pre-trained 2D foundation models into our 3D representations, we unlock powerful open-vocabulary understanding, allowing scenes to be interactively segmented with simple text or clicks without relying on scarce 3D supervised data.\nFirst, I will demonstrate how we can build a queryable 3D representation of large-scale static environments using Lang3D-XL, which enables interactive, text-based exploration by augmenting 3D Gaussian Splatting with low-dimensional semantic features. Building on this, I will extend the concept to dynamic scenes with Dynamic 3D Gaussian Distillation (DGD). This method reconstructs a full 4D semantic scene from a single monocular video, allowing for instantaneous, open-vocabulary segmentation of any object or agent. This enables, for instance, fine-grained interaction and analysis within complex, unstructured 4D data. Finally, I will reverse this information flow with &ldquo;Splat and Distill,&rdquo; showing how 3D reconstruction can serve as a geometric teacher to instill robust 3D awareness back into 2D foundation models.\nSagie Benaim is an Assistant Professor (Senior Lecturer) at the School of Computer Science and Engineering at the Hebrew University of Jerusalem. Previously, he was a postdoc at Copenhagen University, working with Prof. Serge Belongie and as a member of the Pioneer Center for AI. Prior to that, he completed his PhD at Tel Aviv University in the Deep Learning Lab under the supervision of Prof. Lior Wolf. His research interests lie in computer vision, machine learning, and computer graphics, with a particular focus on generative models, neural-based signal representations, and inverse graphics.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:506, Zisapel Building
UID:eventx6a5a287eed47d10877
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20251125T153000
DTEND;TZID=Asia/Jerusalem:20251125T163000
DTSTAMP;TZID=Asia/Jerusalem:20251125T153000
SUMMARY: MSC  talk by Eden Konopnicki  Succinct Zero-knowledge Proofs  at 2025-11-25 15:30:00
DESCRIPTION:Zero-knowledge proofs enable verifying the correctness of computations without revealing any additional information beyond their validity. We focus on proofs that are statistically sound - ensuring that even an unbounded prover cannot convince the verifier of a false statement except with negligible probability, and computationally zero-knowledge. In this work, we study the communication complexity of such proofs. While the original constructions have a large polynomial communication overhead, later works have shown that this overhead can often be significantly reduced.\nWe show that every NP relation that can be verified by a bounded-depth, polynomial-size circuit or a bounded-space, polynomial-time algorithm have a computational zero-knowledge proof whose communication is only additively larger than the witness length. Our construction relies solely on the minimal assumption that one-way functions exist. Moreover, in some cases we achieve the same while making only a black-box use of the one-way function.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 014 &amp; Zoom
UID:eventx6a5a287eed49610865
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20251126T113000
DTEND;TZID=Asia/Jerusalem:20251126T123000
DTSTAMP;TZID=Asia/Jerusalem:20251126T113000
SUMMARY: MSC  talk by Itay Flam  Communication Abstractions for Optimal Byzantine Resilience  at 2025-11-26 11:30:00
DESCRIPTION:We study communication abstractions for asynchronous Byzantine fault tolerance with optimal failure resilience, where n &gt; 3f. Two classic patterns&mdash;canonical asynchronous rounds and communication-closed layers&mdash;have long been considered as general frameworks for designing distributed algorithms, making asynchronous executions appear synchronous and enabling modular reasoning. We show that these patterns are inherently limited in the critical resilience regime 3f &lt; n &le;5f. Several key tasks&mdash;such as approximate and crusader agreement, reliable broadcast and gather&mdash;cannot be solved by bounded-round canonical-round algorithms, and are unsolvable if communication closure is imposed. These results explain the historical difficulty of achieving optimal-resilience algorithms within round-based frameworks. On the positive side, we show that the gather abstraction admits constant-time solutions with optimal resilience (n &gt; 3f ), and supports modular reductions. Specifically, we present the first optimally-resilient algorithm for binding connected consensus by reducing it to gather. For completeness, we present an optimally-resilient algorithm for gather which terminates in constant time. Our results demonstrate that while round-based abstractions are analytically convenient, they obscure the true complexity of Byzantine fault-tolerant algorithms. Richer communication patterns such as gather provide a better foundation for modular, optimal-resilience design.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 601
UID:eventx6a5a287eed4aa10872
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20251126T130000
DTEND;TZID=Asia/Jerusalem:20251126T140000
DTSTAMP;TZID=Asia/Jerusalem:20251126T130000
SUMMARY: Theory Semina  talk by Shay Sapir (Weizmann Institute)  Theory Seminar: Dimension Reduction for Clustering: The Curious Case of Discrete Centers  at 2025-11-26 13:00:00
DESCRIPTION:The Johnson-Lindenstrauss transform is a fundamental method for dimension reduction in Euclidean spaces, that can map any dataset of $n$ points into dimension $O(\log n)$ with low distortion of their distances. This dimension bound is tight in general, but one can bypass it for specific problems. Indeed, tremendous progress has been made for clustering problems, especially in the \emph{continuous} setting where centers can be picked from the ambient space $\mathbb{R}^d$. Most notably, for $k$-median and $k$-means, the dimension bound was improved to $O(\log k)$ [Makarychev, Makarychev and Razenshteyn, STOC 2019].\nWe explore dimension reduction for clustering in the \emph{discrete} setting, where centers can only be picked from the dataset, and present two results that are both parameterized by the doubling dimension of the dataset, denoted as $\operatorname{ddim}$. The first result shows that dimension $O_{\epsilon}(\operatorname{ddim} + \log k + \log\log n)$ suffices, and is moreover tight, to guarantee that the cost is preserved within factor $1\pm\epsilon$ for every set of centers. Our second result eliminates the $\log\log n$ term in the dimension through a relaxation of the guarantee (namely, preserving the cost only for all approximately-optimal sets of centers), which maintains its usefulness for downstream applications.\nOverall, we achieve strong dimension reduction in the discrete setting, and find that it differs from the continuous setting not only in the dimension bound, which depends on the doubling dimension, but also in the guarantees beyond preserving the optimal value, such as which clusterings are preserved.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 401
UID:eventx6a5a287eed4bf10882
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20251130T140000
DTEND;TZID=Asia/Jerusalem:20251130T150000
DTSTAMP;TZID=Asia/Jerusalem:20251130T140000
SUMMARY: MSC  talk by Yitzchak Grinboim  Coverage Models: Monotonicity, and DNA-Sequencing  at 2025-11-30 14:00:00
DESCRIPTION:We study coverage processes where each draw reveals a ℓ-sized subset of [𝑛]. Motivated by DNA-sequencing, we analyze linear and cyclic window models, derive exact and asymptotic results, compare all uniform ℓ-regular models, prove bounds, and conjecture monotonicity while proving the conjecture for broad families of special models.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 601
UID:eventx6a5a287eed4f010881
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20251202T113000
DTEND;TZID=Asia/Jerusalem:20251202T123000
DTSTAMP;TZID=Asia/Jerusalem:20251202T113000
SUMMARY: pixel-club  talk by Nimrod Shabtay (Tel-Aviv University)  Pixel Club: Overcoming Critical Challenges - Towards Reliable, Enhanced, and Efficient VLMs  at 2025-12-02 11:30:00
DESCRIPTION:Vision-language models have achieved impressive performance across diverse tasks, yet they face critical challenges, for example in how we evaluate them, how they use visual context, and how efficiently they process information. This talk presents three interconnected works addressing these limitations. First, LiveXiv provides contamination-free evaluation by automatically generating benchmarks from newly published scientific papers, revealing that some reported VLM improvements may stem from test set contamination rather than genuine advances. Second, IPLoc exposes a surprising gap: current VLMs, struggle with personalized object localization, failing to learn from visual examples the way humans naturally do. By teaching models to focus on contextual cues rather than relying solely on prior knowledge, we significantly improve their few-shot localization abilities. Finally, CARES addresses efficiency by recognizing that not all queries need high-resolution images. Using a lightweight module to predict the minimal sufficient resolution per query, we reduce computational costs by up to 80% while maintaining accuracy. Together, these works demonstrate that context-awareness&mdash;in evaluation, visual reasoning, and resource allocation &ndash; is essential for building VLMs that are more reliable, capable, and practical for real-world deployment\nNimrod Shabtay is a PhD candidate at the faculty of engneering at Tel-Aviv University and a research intern at IBM-Research, supervised by Prof. Raja Giryes.His research focuses on Large Multimodal Models (LMMs). He is particularly interested in overcoming critical challenges towards reliable, enhanced, and efficient LMMs
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:506, Zisapel Building
UID:eventx6a5a287eed50110885
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20251203T103000
DTEND;TZID=Asia/Jerusalem:20251203T113000
DTSTAMP;TZID=Asia/Jerusalem:20251203T103000
SUMMARY: ceClub  talk by Shir Cohen - Researcher at Gen Labs  CE-Club: Bridging Theory and Practice in Distributed Systems  at 2025-12-03 10:30:00
DESCRIPTION:Distributed systems must be worthy of the trust we place in them, yet theory and practice often speak different languages. My research bridges this gap by using theoretical rigor to solve practical problems and real-world deployments to surface new theoretical questions.\nIn this talk, I will present work spanning theoretical and practical aspects of distributed systems. I developed the first sub-quadratic asynchronous Byzantine Agreement algorithm, showing that consensus can be achieved with near-linear communication even without timing assumptions - resolving a long-standing open question. I will also discuss recent work on building and analyzing distributed systems at scale. Together, these projects illustrate how theoretical insights and practical demands can inform each other in the design of reliable distributed infrastructures.\nShort bio:&nbsp;Shir Cohen is a researcher at Gen Labs working on scalable distributed systems. She completed her PhD in Computer Science at Technion under the supervision of Prof. Idit Keidar, where she focused on Byzantine fault tolerance and efficient consensus protocols, and was subsequently a postdoctoral researcher at Cornell University with Prof. Lorenzo Alvisi in the Systems Lab. Her research bridges theory and practice in distributed systems.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zisapel 506
UID:eventx6a5a287eed51710887
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20251203T143000
DTEND;TZID=Asia/Jerusalem:20251203T153000
DTSTAMP;TZID=Asia/Jerusalem:20251203T143000
SUMMARY: MSC  talk by Yuval Shapira  Neural Network Robustness Verification of Few-Pixel Attacks  at 2025-12-03 14:30:00
DESCRIPTION:While successful, neural networks have been shown to be vulnerable to adversarial attacks. In l_0 adversarial attacks, also known as few-pixel attacks, the attacker picks t pixels from the image and arbitrarily perturbs them. While many verifiers prove robustness against l_p attacks for a positive integer p, very little work deals with robustness verification for l_0 attacks. This verification introduces a combinatorial challenge because the space of pixels to perturb is discrete and of exponential size. In this talk, we present a series of papers tackling this challenging problem. We first show that l_&infin; verifiers can be used for l_0 verification, and that by relying on covering designs we can significantly reduce the number of l_&infin; tasks that need to be submitted to the underlying l_&infin; verifier.\nThis idea is implemented in Calzone, the first sound and complete l_0 verifier. We then present CoVerD, which improves upon Calzone by tailoring effective but analysis-incompatible coverings to l_0 robustness verification. Lastly, we characterize the convex hull of the non-convex l_0 perturbation space. Equipped with this geometric perspective, we examine different approaches for linear bound propagation for l_0 verification, which enables us to improve the precision of CoVerD&rsquo;s underlying verifier in our settings and boost its performance.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 301 &amp; Zoom
UID:eventx6a5a287eed52a10884
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20251208T103000
DTEND;TZID=Asia/Jerusalem:20251208T113000
DTSTAMP;TZID=Asia/Jerusalem:20251208T103000
SUMMARY: MSC  talk by Ron Raphaeli  SILO: Solving Inverse Problems with Latent Operators  at 2025-12-08 10:30:00
DESCRIPTION:Plug-and-play methods for solving inverse problems have continuously improved over the years by incorporating more advanced image priors.&nbsp;Latent diffusion models are among the most powerful priors, making them a natural choice for solving inverse problems. &nbsp;\nHowever, existing approaches require multiple applications of an Autoencoder to transition between pixel and latent spaces during restoration, leading to high computational costs and degraded restoration quality. &nbsp;\nIn this work, we introduce a new plug-and-play paradigm that operates entirely in the latent space of diffusion models. By emulating pixel-space degradations directly in the latent space through a short learning phase, we eliminate the need for the Autoencoder during restoration, enabling faster inference and improved restoration fidelity.\nWe validate our method across various image restoration tasks and datasets, achieving significantly higher perceptual quality than previous methods while being 2.6-10 times faster in inference and 1.7-7 times faster when accounting for the learning phase of the latent operator.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 401
UID:eventx6a5a287eed53d10886
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20251208T143000
DTEND;TZID=Asia/Jerusalem:20251208T153000
DTSTAMP;TZID=Asia/Jerusalem:20251208T143000
SUMMARY: MSC  talk by Shachar Cohen  Guiding and Assisting the Breast Cancer Treatment Process Through Analysis of Whole Slide Images With Deep Learning  at 2025-12-08 14:30:00
DESCRIPTION:Breast cancer treatment decisions heavily rely on biomarker assessments, traditionally obtained through resource-intensive chemical processes and genomic assays. This introduces challenges, including high costs, long turnaround times, and inter-observer variability. Additionally, these methods may be unavailable in some countries. This work explores deep learning-based analysis of gigapixel whole slide images (WSIs) to predict biomarker expression, guide and assist the treatment process, focusing on hematoxylin and eosin (H&amp;E)-stained slides as a cost-effective alternative. We employed multiple instance learning (MIL) frameworks, progressed from attention-based MIL to transformer-based MIL and vision transformers, leveraging inter-patch relationships in WSIs to handle slide-level labels effectively. Models were trained and validated on large multi-institutional cohorts, demonstrating robust performance with a high area under the ROC curve (AUC) for certain biomarker predictions (0.93&ndash;0.96 across independent cohorts). For recurrence risk, integration of the Prov-GigaPath foundation model achieved strong OncotypeDX score discrimination (external AUC of 0.82) with excellent generalization. Clinical utility was rigorously evaluated through sensitivity-specificity trade-offs, negative predictive value (NPV) for low-risk identification (21&ndash;25% of patients with NPV 0.96&ndash;0.97), and survival analysis, yielding significant hazard ratios (2.0&ndash;4.1) for recurrence-free and breast cancer-specific survival. Following these evaluations, the results highlight the potential of deep learning on WSIs to spare unnecessary testing for specific patient groups, enhancing accessibility, reducing costs, and refining personalized treatment strategies&mdash;such as guiding chemotherapy decisions and forecasting recurrence risk&mdash;ultimately paving the way for the clinical integration of AI-assisted pathology.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 401
UID:eventx6a5a287eed54f10879
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20251210T130000
DTEND;TZID=Asia/Jerusalem:20251210T140000
DTSTAMP;TZID=Asia/Jerusalem:20251210T130000
SUMMARY: Theory Semina  talk by Yaseen Abd-Elhaleem (University of Haifa)  Theory Seminar: Distributed Maximum Flow in Planar Graphs  at 2025-12-10 13:00:00
DESCRIPTION:The dual of a planar graph G is a planar graph G* that has a vertex for each face of G and an edge for each pair of adjacent faces of G. The profound relationship between G and G* is an algorithmic basis for solving numerous (centralized) classical problems on planar graphs. In the standard distributed CONGEST model however, the use of planar duality is very restricted.\nWe extend the distributed algorithmic toolkit to work on G*, facilitating various algorithms for classical problems on G. E.g., state-of-the-art algorithms for Maximum st-Flow and Undirected Minimum Weight Cycle.\nBased on a joint work with Michal Dory, Merav Parter and Oren Weimann.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 401
UID:eventx6a5a287eed56010892
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20251211T143000
DTEND;TZID=Asia/Jerusalem:20251211T153000
DTSTAMP;TZID=Asia/Jerusalem:20251211T143000
SUMMARY: MSC  talk by Asaf Shoham  Forest Factorization of Forests  at 2025-12-11 14:30:00
DESCRIPTION:Simon's Factorization Theorem is a powerful tool in the study of formal languages, and is key to many algorithms and theoretical results.\nThe theorem states that, given a regular language, we can decompose any word to a bounded-height tree that allows reasoning about all infixes of the word in a compact manner.&nbsp;\nIn this work, we generalize Simon's factorization theorem to languages of&nbsp;trees. This generalization brings about many challenges involving the manipulation of trees and forests. Specifically, even defining what a decomposition of trees means is highly nontrivial. We provide such a notion, and show how to obtain it (with a caveat).
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Room Taub 601 &amp;&nbsp;Zoom
UID:eventx6a5a287eed56e10890
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20251214T103000
DTEND;TZID=Asia/Jerusalem:20251214T113000
DTSTAMP;TZID=Asia/Jerusalem:20251214T103000
SUMMARY: MSC  talk by Malak Marrid  Automatic Abstraction Refinement for Hyperproperties Verification  at 2025-12-14 10:30:00
DESCRIPTION:Hyperproperties specify the behavior of a system across multiple executions, and are an important extension of regular temporal properties. Most algorithms for deciding if a given system satisfy a given&nbsp;hyperproperty rely on a user-specified abstraction of the system. In this work, we suggest a novel automatic abstraction-refinement algorithm for hyperproperties verification. Our approach is based on predicate abstraction and the recently introduced reduction of hyperproperties verification to satisfiability of Constrained Horn Clauses (CHCs).\nMoreover, it formalizes and uses CHC-based refinement for counterexamples in the shape of a directed acyclic graph. We implemented our new algorithm on top of the SMT solver Z3. Our experimental evaluation shows our automatic abstraction refinement algorithm can solve a variety of hyperproperty verification problems, completely automatically. This is in contrast to other existing techniques that require a user-given abstraction.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 601
UID:eventx6a5a287eed57e10891
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20251216T113000
DTEND;TZID=Asia/Jerusalem:20251216T133000
DTSTAMP;TZID=Asia/Jerusalem:20251216T113000
SUMMARY: pixel-club  talk by Leo Segre (Tel-Aviv University)  Pixel Club: Understanding Scenes as 3D-Consistent Representations  at 2025-12-16 11:30:00
DESCRIPTION:In this talk, we explore methods for understanding and manipulating 3D scenes through consistent geometric and photometric representations. We begin with VF-NeRF, an approach for NeRF registration that aligns scenes using visibility-aware novel views. We then describe Optimize the Unseen, a method that leverages a free-space prior to improve NeRF reconstructions by removing artifacts in regions with limited observations. Next, we introduce a frequency-aware decomposition for 3D Gaussian Splatting, enabling progressive rendering, foveated visualization, and efficient interaction with complex scenes. Finally, we present Multi-View Foundation Models, which incorporate multi-view consistency into vision foundation models to produce 3D-aware representations directly from 2D features.Together, these contributions highlight how visibility, frequency structure, and multi-view reasoning can lead to more expressive and reliable 3D scene representations.\nLeo Segre is a PhD candidate at Tel Aviv University, supervised by Prof. Shai Avidan. His research centers on understanding how 3D structure, visibility, and multi-view relationships can be used to improve learned representations. He works on neural scene representations and 3D-aware vision models, with an emphasis on algorithms that combine geometric constraints with data-driven learning.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:506, Zisapel Building
UID:eventx6a5a287eed58e10893
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20251222T103000
DTEND;TZID=Asia/Jerusalem:20251222T113000
DTSTAMP;TZID=Asia/Jerusalem:20251222T103000
SUMMARY: pixel-club  talk by Meir Yossef Levi (Technion)  Pixel Club: Understanding CLIP Latent Space - Where the Common and Rare Images Reside?  at 2025-12-22 10:30:00
DESCRIPTION:CLIP is a pioneering work for embedding image and text into a shared latent space using contrastive learning. It facilitates semantic space which propelled wide range of vision tasks, from retrieval, classification to text-to-image synthesis.\nIn this seminar I will focus on to the geometry of CLIP&rsquo;s latent space, drawing on two ICML 2025 papers.\n\n\nThe latent space of CLIP is actually modeled by two shifted non-isometric ellipsoids; one for images and one for text, rather than a shared hypersphere. This perspective uncovers systematic geometric biases and motivates conformity, a measure capturing how common or rare a concept is, and where geometrically it resides. Common concepts exhibit high conformity, while rare ones lie farther from the mean, revealing a geometric view of concept commonality.\n\n\nThen, a simple whitening transformation further maps the latent space into an isotropic form where embedding norms correlate with likelihood. This enables practical applications such as OOD detection and identifying generative artifacts, offering a cohesive geometric&ndash;probabilistic understanding of CLIP.\n\nI will begin with a short brief on my earlier works on robust 3D classification, highlighting how robustness analysis and explainability techniques reveal the structural behavior of point-cloud classifiers.\n\n\nThis is a seminar talk for PhD candidacy of Meir Yossef Levi under the supervision of Prof. Guy Gilboa.\n\n\nMeir Yossef Levi (Yossi Levi) is in the final stages of his Ph.D. at the Technion, advised by Prof. Guy Gilboa, after receiving both his B.Sc. and M.Sc. in Electrical Engineering from the Technion. His research focuses on multimodal representation learning, with a particular interest in understanding the latent geometry of vision-language models and its implications. His recent work centers on CLIP, with two papers accepted to ICML 2025 on this topic. Prior to this, he studied robust classification in 3D vision, with publications at ICCV and 3DV.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:506, Zisapel Building
UID:eventx6a5a287eed59e10897
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20251223T113000
DTEND;TZID=Asia/Jerusalem:20251223T123000
DTSTAMP;TZID=Asia/Jerusalem:20251223T113000
SUMMARY: pixel-club  talk by Yair Weiss (Hebrew University of Jerusalem)  Pixel Club: What makes deep generative models of images work?  at 2025-12-23 11:30:00
DESCRIPTION:Perhaps the most mysterious aspect of modern deep generative models of images is that they work even when the number of training examples is much smaller than the dimensionality of the input. Often this is attributed to the &ldquo;manifold hypothesis&rdquo; which argues that the models estimate a low dimensional manifold that best fits the training distribution, but I will show that this explanation is flawed. Rather I will present theoretical and empirical results which demonstrate that architectural choices made in successful GANs and diffusion models make them learn the distribution of patches rather than the distribution of images. Finally, I will show work in progress where we apply this insight (&ldquo;patches are all you need&rdquo;) to classical methods for image generation.\nJoint work with: Ariel Elnekave, Roy Friedman, Itamar Harel and Antonio Torralba.Yair Weiss is the Dieter Schwarz Professor of Artificial Intelligence at the Hebrew University and the former Dean of the School of Computer Science and Engineering. His research interests include Human and Machine Vision, Machine Learning and Neural Computation.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:1061, Meyer Building
UID:eventx6a5a287eed5b310898
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20251223T180000
DTEND;TZID=Asia/Jerusalem:20251223T200000
DTSTAMP;TZID=Asia/Jerusalem:20251223T180000
SUMMARY: CSpecial Event  AI and Chip Design Event with Intel  at 2025-12-23 18:00:00
DESCRIPTION:Join us for an event that will address challenges in chip design and addressing them with the help of artificial intelligence (AI).\nTuesday, 12/23, starting at 6:00 PM at the Technion - Meyer Building 815\nThe event will include a lecture and roundtables with Intel engineers, to discuss ideas for future AI-based projects&nbsp;and of course, a march and other surprises!\nPlease register in advance here (limited seats)
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Meyer Building 815
UID:eventx6a5a287eed5c310894
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20251224T103000
DTEND;TZID=Asia/Jerusalem:20251224T113000
DTSTAMP;TZID=Asia/Jerusalem:20251224T103000
SUMMARY: MSC  talk by Itay Segev  Multi-Objective Risk-Aware Actor–Critic  at 2025-12-24 10:30:00
DESCRIPTION:Real-world reinforcement learning must optimize multiple objectives, satisfy safety constraints, and remain robust to model uncertainty, yet existing methods tackle these challenges in isolation. We introduce Multi-Perspective Actor&ndash;Critic (MPAC), a unified framework that combines value decomposition with component-specific risk assessment, enabling safety-critical objectives to act conservatively while performance-oriented objectives retain appropriate optimism.\nAn influence-based weighting mechanism dynamically adjusts objective importance based on decision relevance and learning progress, eliminating the need for fixed scalarization or prior reward tuning. This produces policies that are simultaneously safe, robust to perturbations, and less conservative than traditional safe or robust RL approaches.\nEvaluation on a complex energy-management domain, together with continuous-control benchmarks featuring safety constraints and perturbed dynamics, shows that MPAC consistently achieves stronger multi-objective trade-offs.&nbsp;
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 601
UID:eventx6a5a287eed64710888
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20251224T113000
DTEND;TZID=Asia/Jerusalem:20251224T123000
DTSTAMP;TZID=Asia/Jerusalem:20251224T113000
SUMMARY: ceClub  talk by Dr. Jacob Leshno (University of Chicago Booth)  CE-Club: On the Viability of Open-Source Financial Rails: Economic Security of Permissionless Consensus (Joint work with Rafael Pass and Elaine Shi)  at 2025-12-24 11:30:00
DESCRIPTION:Bitcoin showed the possibility of an open-source financial rail, relying on interchangeable service providers to join or leave at will. A key question is whether the permissionless consensus technology that underpins this permissionless design can also meet the security standards required for mainstream financial applications. We develop a model that integrates economic and distributed-systems constraints, define the objectives of free entry and security, and examine their joint attainability. We demonstrate the feasibility of an open and secure protocol by presenting a protocol that attains an economically meaningful notion of security while preserving Bitcoin&rsquo;s permissionless design. Our protocol&rsquo;s security does not require costly miner rewards or high energy consumption. The analysis formalizes the essential role of the user community in sustaining secure and efficient open financial rails.&nbsp;paper link
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:
UID:eventx6a5a287eed69110901
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20251224T130000
DTEND;TZID=Asia/Jerusalem:20251224T140000
DTSTAMP;TZID=Asia/Jerusalem:20251224T130000
SUMMARY: Theory Semina  talk by Yuval Gil (Reykjavik University)  Theory Seminar: Distributed Interactive Proofs for Planarity with Log-Star Communication  at 2025-12-24 13:00:00
DESCRIPTION:The notion of a distributed interactive proof (DIP) was introduced by Kol, Oshman, and Saxena (PODC 2018). In a DIP, the verifier seeks to verify a certain claim regarding a given input graph $G$ in a distributed fashion, i.e., operating concurrently on all $n$ nodes of $G$. The verification is done by interacting with a centralized prover that has access to the entire input graph. A DIP is measured by the amount of communication it requires. Namely, the objective is to design DIPs with a small number of interaction rounds and a small proof size, i.e., a small message size per interaction round.&nbsp;We focus on the planarity task, i.e., deciding if the input graph is planar. Naor, Parter, and Yogev (SODA 2020) present a DIP for planarity with $3$ interaction rounds and a proof size of $O(\log n)$. Shortly after, Feuilloley et al. (PODC 2020) showed that the same proof size can be accomplished by a non-interactive protocol and gave a matching lower bound for the non-interactive case. In this talk, I will discuss two recent papers (DISC 2025, SODA 2026) that provide new DIP protocols with significantly improved communication bounds culminating in a protocol with only $O(\log ^{*} n)$ communication. The talk will be self-contained --- no prior knowledge on planarity/distributed interactive proofs is necessary.&nbsp;Based on joint work with Merav Parter.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 401
UID:eventx6a5a287eed6ac10899
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20251228T180000
DTEND;TZID=Asia/Jerusalem:20251228T200000
DTSTAMP;TZID=Asia/Jerusalem:20251228T180000
SUMMARY: CSpecial Event  "Entrepreneurship on the Bar" Night  at 2025-12-28 18:00:00
DESCRIPTION:Join an event of inspiration, beer and technology!Two of our faculty members - Prof. Ran El-Yaniv and Prof. Yoav Etsion, talk about their personal entrepreneurial journey that grew from academic research, became a successful startup and reached the technology giants\nWhen: Sunday 12/28 at 6:00 PM at Taub 2To register (limited number of places): here\nOn the program:\nEntrepreneurship, Academia and Deep LearningProf. Ran El-Yaniv, Professor of Computer Science at the Technion and Senior Director at Nvidia, Founding Partner of Deci AI which was acquired by Nvidia in 2024.Deep learning, optimization of neural networks - this is the basis of Deci AI! Prof. El-Yanib will present the key technological elements in building the startup that was sold to Nvidia, and delve into his personal story. What are the practical lessons from the startup journey? How do you maintain the connection between groundbreaking academic research and technological entrepreneurship?\nFrom academia to hardware accelerator company: on data analysis, silicon, and peopleProf. Yoav Etsion, Professor of Computer Science and Electrical and Computer Engineering at the Technion and Chief Technologist at Speedata.The world is flooded with information, and data centers are struggling to meet computing and energy demands. Why do we need hardware accelerators instead of regular processors? Which industries are facing the challenge, and why are neural networks not the solution?Along the way, we will learn about the journey from academia to founding a chip company.\nWe are waiting for you with beers and pizzas!
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 2
UID:eventx6a5a287eed6bf10895
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20251229T103000
DTEND;TZID=Asia/Jerusalem:20251229T113000
DTSTAMP;TZID=Asia/Jerusalem:20251229T103000
SUMMARY: MSC  talk by Itay Hasson  Box-Reachability in Vector Addition Systems  at 2025-12-29 10:30:00
DESCRIPTION:We consider a variant of reachability in Vector Addition Systems (VAS) dubbed box reachability, whereby a vector v in N^d is box-reachable from 0 in a VAS V if V admits a path from 0 to v that not only stays in the positive orthant (as in the standard VAS semantics), but also stays below v, i.e., within the ׳׳box׳׳ whose opposite corners are 0 and v.\nOur main result is that for two-dimensional VAS, the set of box-reachable vertices almost coincides with the standard reachability set: the two sets coincide for all vectors whose coordinates are both above some threshold W. We also study properties of box-reachability, exploring the differences and similarities with standard reachability.\nTechnically, our main result is proved using powerful machinery from convex geometry.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 601
UID:eventx6a5a287eed6d310874
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20251229T133000
DTEND;TZID=Asia/Jerusalem:20251229T143000
DTSTAMP;TZID=Asia/Jerusalem:20251229T133000
SUMMARY: TDC Seminar  talk by Yarden Adi (Technion)  TDC Seminar: Communication-Efficient and Fast Local Symmetry Breaking  at 2025-12-29 13:30:00
DESCRIPTION:We study the cost of communication in designing fast distributed graph algorithms.\nFor many graph problems, the phrase ``fast distributed algorithms'' refers to algorithms whose running time is $\polylog(n)$ (or even faster).Problems that admit such fast algorithms are usually referred to as local problems and include fundamental symmetry breaking problems such as maximal independent set (MIS), $(\Delta + 1)$-coloring, and maximal matching (MM), which have been studied extensively for several decades.\nHowever, the communication cost of all known fast algorithms for fundamental local symmetry breaking problems is high, in general, as high as $\Omega(n^{2})$;in fact, for the aforementioned problems, there are impossibility results, showing that communicating $\Omega(n^{2})$ bits is, in general, mandatory, regardless of running time.\nThis raises the following cardinal question:Do there exist distributed algorithms for important local symmetry breaking problems that are simultaneously communication-efficient and fast?\nWe answer this question in the affirmative by developing fast distributed algorithms with near-linear (in $n$) communication complexity for two ruling set problems, which are important generalizations of MIS that have been extensively studied and widely used in distributed computing.Specifically, the main contributions of this work are randomized distributed algorithms that run for $O(\log n)$ rounds, send $n \polylog(n)$ constant size messages, and construct $(\alpha, \beta)$-ruling sets with high probability for$(\alpha, \beta) \in \{ (2, 3), (3, 4) \}$.\nWe emphasize that these algorithms are near-optimal in terms of both their communication complexity and round complexity (using constant size messages).Our ruling set constructions are based on the design of a communication- and time-efficient distributed algorithm for a new graph theoretic problem, called low volume $k$-dominating set, which may be of independent interest.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Meyer 1061
UID:eventx6a5a287eed6e510903
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20251230T103000
DTEND;TZID=Asia/Jerusalem:20251230T113000
DTSTAMP;TZID=Asia/Jerusalem:20251230T103000
SUMMARY: colloq  talk by Roni Con  Correcting Deletions: Fundamental Questions and Surprising Applications  at 2025-12-30 10:30:00
DESCRIPTION:Error-correcting codes are central to information theory and theoretical computer science, enabling reliable communication in the presence of noise. Classical results focus primarily on substitutions and erasures, and over the years elegant constructions have emerged that meet, or closely approach, the optimal rate&ndash;noise tradeoffs in these settings. A natural and equally fundamental error model involves synchronization errors, such as insertions and deletions. These errors, already studied in the 1960s, cause misalignment between sender and receiver and arise in various modern technologies, with DNA-based data storage being a particularly compelling example. Despite decades of attention, fully characterizing the capacity of synchronization channels and constructing practical, near-optimal codes for them remain major open challenges.In this talk, I will survey recent progress on coding for synchronization channels and then focus on the performance of linear and Reed&ndash;Solomon codes under insertions and deletions, demonstrating that well-structured algebraic codes can, perhaps unexpectedly, correct a significant amount of synchronization noise. I will then highlight a surprising application of these ideas in secret sharing. Finally, I will discuss how input-correlated insertion/deletion channels naturally arise in DNA-based data storage, an emerging ultra-dense archival technology, and present capacity theorems and efficient coding schemes tailored to an important class of such channels.\nShort Bio: Roni Con is a postdoctoral researcher at the Technion, hosted by Prof. Eitan Yaakobi. In Spring 2024, he was a Simons Research Fellow in the program Error-Correcting Codes: Theory and Practice at the Simons Institute for the Theory of Computing. He completed his Ph.D. in October 2023 at Tel Aviv University under the supervision of Profs. Amir Shpilka and Zachi Tamo. His research focuses on error-correcting codes and information theory, with applications to synchronization-error channels, modern storage systems, and cryptography.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Auditorium 012, Floor 0
UID:eventx6a5a287eed94a10902
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20251230T113000
DTEND;TZID=Asia/Jerusalem:20251230T123000
DTSTAMP;TZID=Asia/Jerusalem:20251230T113000
SUMMARY: cggc  talk by Prof. Amir Vaxman (The School of Informatics, The University of Edinburgh)  CGGC Seminar: Geometric Representations Matter  at 2025-12-30 11:30:00
DESCRIPTION:I will discuss recent work highlighting the importance of selecting an appropriate geometric representation for tasks such as 3D/4D generation, medical imaging, and motion capture. &nbsp;I will argue that this choice should follow directly from the problem&rsquo;s inherent constraints and the desired outcomes.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Auditorium 012, Taub building floor 0
UID:eventx6a5a287eedbad10896
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20251230T113000
DTEND;TZID=Asia/Jerusalem:20251230T123000
DTSTAMP;TZID=Asia/Jerusalem:20251230T113000
SUMMARY: pixel-club  talk by Tom Tirer (Bar Ilan University)  Pixel Club: Low-Level Processing: Provable Benefits for High-Level Tasks and a Zero-Shot Restoration Framework That Works with as Little as 4 Iterations  at 2025-12-30 11:30:00
DESCRIPTION:The data processing inequality is an information-theoretic principle stating that the information content of a signal cannot be increased by processing the observations. In particular, it suggests that there is no benefit in enhancing the signal or encoding it before addressing a classification problem. This assertion can be proven to be true for the case of the optimal Bayes classifier. However, in practice, it is common to perform &ldquo;low-level&rdquo; tasks before &ldquo;high-level&rdquo; downstream tasks despite the overwhelming capabilities of modern deep neural networks. With the aim of understanding when and why low-level processing can be beneficial for classification, I will present a comprehensive theoretical study of a binary classification setup, where we consider a classifier that is tightly connected to the optimal Bayes classifier and converges to it as the number of training samples increases. We prove that for any finite number of training samples, there exists a pre-classification processing that improves the classification accuracy. We explore the factors that affect this gain and discover a non-intuitive relation between the maximal gain and a notion of signal-to-noise ratio. Importantly, we conduct an empirical study where we investigate the effect of denoising and encoding on the performance of practical deep classifiers on benchmark datasets and demonstrate trends that are consistent with our theoretical results.\nMotivated by the first part of the talk, which shows the usefulness of low-level tasks, in the second part I will focus on &ldquo;zero-shot&rdquo; (training-free) image restoration using the guidance of Diffusion Models (DMs) and their variants. We hypothesize that current methods require many Neural Function Evaluations (NFEs) for performing well due to the many NFEs needed in the original generative functionality of the DMs. On the other hand, existing restoration methods that use faster generative models, such as Consistency Models (CMs), still require tens of NFEs or fine-tuning of the model per task. To fill this gap, I will present a zero-shot restoration scheme that uses CMs and operates well with as little as 4 NFEs. It is based on a wise combination of several ingredients: better initialization, back-projection guidance, and, above all, a novel noise injection approach that acts as momentum. We demonstrate the advantages of our approach for image super-resolution, deblurring and inpainting. Interestingly, we show that the usefulness of our noise injection strategy goes beyond CMs: it can also mitigate the performance degradation of existing guided DM methods when reducing their NFE count.\nTom Tirer is a Senior Lecturer (Assistant Professor) in the Faculty of Engineering at Bar-Ilan University. He received his BSc degree (summa cum laude) in electrical engineering from Ben-Gurion University of the Negev in 2010, and his MSc (summa cum laude) and PhD degrees in electrical engineering from Tel Aviv University in 2016 and 2020, where he also was a postdoctoral researcher during 2021. In 2022 he was a postdoctoral researcher at NYU Center for Data Science. His research interests are in the (often intersecting) fields of signal and image processing, machine learning and optimization. Alongside his academic endeavors, he also worked for several years in the industry in various engineering, algorithms and research roles.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:1061, Meyer Building
UID:eventx6a5a287eedbc410906
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20251231T113000
DTEND;TZID=Asia/Jerusalem:20251231T123000
DTSTAMP;TZID=Asia/Jerusalem:20251231T113000
SUMMARY: ceClub  talk by David Trabish (Technion)  CE-Club: Enhancing Symbolic Execution with Machine-Checked Safety Proofs  at 2025-12-31 11:30:00
DESCRIPTION:Symbolic execution (SE) is a program analysis technique that executes the program with symbolic inputs. In modern SE engines, when the analysis of a given program is exhaustive, the analyzed program is typically considered safe, i.e., free of bugs, but no formal guarantees are provided to support this. Rather than aiming for a formally verified SE engine that will provide such guarantees, which is challenging, we propose a systematic approach where each individual analysis additionally outputs a formal safety proof that validates the symbolic computations that were carried out.\nOur approach consists of two main components: A formal framework connecting concrete and symbolic semantics, and an instrumentation of the SE engine that generates formal safety proofs based on this framework. We showcase our approach by implementing a KLEE-based prototype that operates on a subset of LLVM IR with integers and generates proofs in Coq.Our preliminary experiments show that our approach generates proofs that have reasonable validation times, while the instrumentation incurs only a minor overhead on the SE engine.\nIn addition, during the implementation of our prototype, we found previously unknown semantic implementation issues in KLEE.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Mayer 1061&nbsp;
UID:eventx6a5a287eedbe210910
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20251231T123000
DTEND;TZID=Asia/Jerusalem:20251231T133000
DTSTAMP;TZID=Asia/Jerusalem:20251231T123000
SUMMARY: MSC  talk by Amit Levi  What I’ve learned so far about the illusion of safety in LLM alignment  at 2025-12-31 12:30:00
DESCRIPTION:As AI capabilities rapidly advance toward AGI, safety and security often lag behind power and profit. A core challenge in AI safety is our inability to reliably evaluate even basic safe behavior in large language models (LLMs), leading to misalignment, fake alignment, unreliable safety benchmarks, and mismeasured risks. My AI safety journey began in 2023 during my second semester of a BSc at the Technion, as an exchange student at EPFL, where graduate-level AI courses led me to first train and safety fine-tune LLMs. By early 2024, I began working with Professor Avi Mendelson on alignment vulnerabilities. In 2024, we showed that an initialization-based jailbreak can effectively reverse a model&rsquo;s safety alignment using only ~20 tokens, demonstrating that current safety training is not truly informative (EMNLP/ACL 2025). Building on this work, we identified a vulnerability caused by alignment training itself: existing bias and fairness evaluations detect at best ~80% of biases, and many debiasing methods create fake alignment rather than addressing underlying issues. We proposed a benchmark that surfaces both visible and hidden biases (oral, AAAI 2026). Following this, we developed three additional works on pad-token influence, failures of unlearning evaluation, and quantization via hidden representations, A2A vulnerabilities (on-going, Black-Hat, NeurIPS 2025),and alongside many more ongoing alignment research we are doing this days. In this seminar, I will present key insights from my research works work on alignment vulnerabilities, adversarial failures, fake alignment, unlearning evaluation failures, and the risks posed by the current trajectory of AI development with current alignment methods.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 601 &amp; Zoom
UID:eventx6a5a287eedbf810907
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20251231T130000
DTEND;TZID=Asia/Jerusalem:20251231T140000
DTSTAMP;TZID=Asia/Jerusalem:20251231T130000
SUMMARY: colloq  talk by Yotam Dikstein  High Dimensional Expanders: Structure and Applications  at 2025-12-31 13:00:00
DESCRIPTION:Expanders are graphs that are both edge-sparse and well connected. They have been an instrumental tool in many results in mathematics and computer science, some of which seem to have little connection to graphs at all. High dimensional expanders (HDXs) are hypergraph analogues of expander graphs &mdash; sparse and well connected hypergraphs.&nbsp; HDXs have already played a key role in several recent results in theoretical computer science and combinatorics. However, much of their underlying theory remains unexplored.\nMotivated by this, we will see what HDXs are, why they generalize expander graphs, and how they serve as a common setting for results in probability and computational complexity.&nbsp;\nI will illustrate their power by presenting new Chernoff-type inequalities for HDXs and explaining their applications to:&nbsp;\n1. Hardness amplification - constructing functions for which the best polynomial-time algorithm performs no better than random guessing.\n2. Hardness of approximation - constructing natural problems for which even approximate solutions are intractable unless P=NP.\nThe talk will not assume prior background and is aimed at a broad audience.&nbsp;\nBio:&nbsp;Yotam is a theoretical computer scientist working between combinatorics and computational complexity. His work studies combinatorial structures such as high-dimensional expanders, and their implications for algorithms and complexity theory. He is currently a postdoc fellow at the Institute for Advanced Study and received his PhD from the Weizmann Institute of Science.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Auditorium 012, Floor 0
UID:eventx6a5a287eedc0f10904
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20260101T123000
DTEND;TZID=Asia/Jerusalem:20260101T133000
DTSTAMP;TZID=Asia/Jerusalem:20260101T123000
SUMMARY: Theory Semina  talk by Aviad Rubinstein (Stanford University)  Theory Seminar: Algorithms for Online Calibration, Edit Distance, and LCS via the "Average Scale is Boring Principle"  at 2026-01-01 12:30:00
DESCRIPTION:I will explain a phenomenon that I call the "Average Scale is Boring Principle" (you may help me come up with a better name): The 01010101... sequence, for example, has high variance on a local scale, but if you look at larger scales each segment looks identical. For another example, the sequence 0^n1^n looks different at a global scale, but on almost all short intervals it is constant. I will formalize a sense in which some sequences have high variance on some scales, but for *all* sequences the average scale is "boring". (Bonus: the proof is only a few lines!)\nInterestingly, the same principle recently came in handy in two very different projects: one on approximation algorithms for Edit Distance and Longest Common Subsequence, and the other on online algorithms for calibrated forecasting.\nThe talk is based on things I learnt from Xiao Mao and Binghui Peng.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 401
UID:eventx6a5a287eedc2410909
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20260104T103000
DTEND;TZID=Asia/Jerusalem:20260104T113000
DTSTAMP;TZID=Asia/Jerusalem:20260104T103000
SUMMARY: MSC  talk by Mayan Rivlin  Data Analysis and DNA Design with Applications  at 2026-01-04 10:30:00
DESCRIPTION:CRISPR-Cas genome editing is a highly specific technology that enables edit operations in genomes and other DNA fragments. CRISPR-Cas as a tool in biotechnology is the result of harnessing a similar system that naturally occurs in bacteria. The technology holds promise as a component in therapy and clinical approaches. Indeed - several therapies were recently approved for use in patients. For example - Casgevy, a sickle cell anemia therapy that is showing great positive results. One of the challenges in using CRISPR-Cas technology is the risk of off-target activity - edit events that can occur in undesired locations in the target genome or DNA mixture. In my talk I will present the results of large-scale analysis designed to identify the sequence determinants of CRISPR-Cas12a off-target activity. The data we analyzed is produced by high throughput DNA synthesis. Using several statistical approaches, including the minimum hyper geometric framework (mHG - developed at the Technion in 2007), we are able to point to characteristics of cleavage efficiency based on hundreds of thousands of variants for which activity was measured. All experimental work was done in collaboration with TUM in Munich. By identifying target sequence determinants, our work aims to improve the safety and precision of CRISPR-based applications.\nAs global data generation grows exponentially, DNA-based data storage technology has emerged as a revolutionary solution due to its high density, stability, and longevity. In the second part of the talk I will describe computational challenges involved in the use of encoding data into composite DNA (work initiated at the Technion in 2016), and solutions to some of these challenges. &nbsp;
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 601
UID:eventx6a5a287eedc3510905
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20260104T103000
DTEND;TZID=Asia/Jerusalem:20260104T113000
DTSTAMP;TZID=Asia/Jerusalem:20260104T103000
SUMMARY: colloq  talk by Yael Vinker  Bridging Generative Models and Visual Communication  at 2026-01-04 10:30:00
DESCRIPTION:From rough sketches that spark ideas to polished designs that explain complex concepts, visual communication is central to how humans think, create, and share knowledge. Yet despite major advances in generative AI, we are still far from models that can reason and communicate through visual forms.I will present my work on bridging generative models and visual communication, focusing on three complementary domains: (1) algorithms for generating and understanding sketches, (2) systems that support exploratory visual creation beyond one-shot generation, and (3) methods for producing editable, parametric images for design applications.\nThese domains pose unique challenges, they are inherently data-scarce and rely on representations that go beyond pixel-based images commonly used in standard models. I will show how the rich priors of vision-language models can be leveraged to address these challenges through novel optimization objectives and regularization techniques that connect their learned features with the specialized representations required for visual communication.Looking ahead, this research lays the foundation for general-purpose visual communication technologies: intelligent systems that collaborate with humans in visual domains, enhancing how we design, learn, and exchange knowledge.\nBio: Yael Vinker is a Postdoctoral Associate at MIT CSAIL, working with Prof. Antonio Torralba. She received her Ph.D. in Computer Science from Tel Aviv University, advised by Profs. Daniel Cohen-Or and Ariel Shamir. Her research spans computer graphics, computer vision, and machine learning, with a focus on generative models for visual communication. Her work has been recognized with two Best Paper Awards (SIGGRAPH 2022, SIGGRAPH Asia 2023) and a Best Paper Honorable Mention (SIGGRAPH 2023). She was selected as an MIT EECS Rising Star (2024) and received the Blavatnik Prize for Outstanding Israeli Doctoral Students in Computer Science (2024) as well as the VATAT Ph.D. Fellowship.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Auditorium 012, Floor 0
UID:eventx6a5a287eedc4b10911
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20260105T103000
DTEND;TZID=Asia/Jerusalem:20260105T113000
DTSTAMP;TZID=Asia/Jerusalem:20260105T103000
SUMMARY: colloq  talk by Gilad Yehudai   Two Lenses on Deep Learning: Data Reconstruction and Transformer Structure  at 2026-01-05 10:30:00
DESCRIPTION:Despite the remarkable success of modern deep learning, our theoretical understanding remains limited. Many fundamental questions about how these models learn, what they memorize, and what their architectures can express are still largely open. In this talk, I focus on two&nbsp;such questions that offer complementary perspectives on the behavior of modern networks.\nFirst, I examine how standard training procedures implicitly encode aspects of the training data in the learned parameters, enabling reconstruction across a wide range of architectures and loss functions. Second, I turn to transformers and analyze how architectural choices, such as the number of heads, rank, and depth, shape their expressive capabilities, revealing both strengths and inherent limitations of low-rank attention.\nTogether, these perspectives highlight recurring principles that shape the behavior of deep models, bringing us closer to a theoretical framework that can explain and predict the phenomena observed in practice.&nbsp;\nBio: Gilad is a postdoctoral research associate at the Courant Institute of Mathematical Sciences at New York University, hosted by Prof. Joan Bruna. His research focuses on the theory of deep learning models, with a recent emphasis on transformers. He also works on attacks on neural networks, particularly data reconstruction attacks and adversarial attacks. Previously, he completed his Ph.D. at the Weizmann Institute of Science under the supervision of Ohad Shamir. He has held research internships at NVIDIA, where he worked on graph neural networks, and at Google Research, where he worked on large-scale optimization. He holds a B.Sc. and M.Sc. in mathematics from Tel Aviv University.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Auditorium 012, Floor 0
UID:eventx6a5a287eedc5f10912
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20260105T133000
DTEND;TZID=Asia/Jerusalem:20260105T143000
DTSTAMP;TZID=Asia/Jerusalem:20260105T133000
SUMMARY: TDC Seminar  talk by Yuval Efron (Columbia University)  Optimal Good-Case Latency for Sleepy Consensus  at 2026-01-05 13:30:00
DESCRIPTION:In the context of Byzantine consensus problems such as Byzantine broadcast (BB) and Byzantine agreement (BA), the good-case setting aims to study the minimal possible latency of a BB or BA protocol under certain favorable conditions, namely the designated leader being correct (for BB), or all parties having the same input value (for BA). We provide a full characterization of the feasibility and impossibility of good-case latency, for both BA and BB, in the synchronous sleepy model. Surprisingly to us, we find irrational resilience thresholds emerging: 2-round good-case BB is possible if and only if at all times, at least 0.618 fraction of the active parties are correct; 1-round good-case BA is possible if and only if at least 0.707 fraction of the active parties are correct.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 601
UID:eventx6a5a287eedc7110915
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20260105T173000
DTEND;TZID=Asia/Jerusalem:20260105T190000
DTSTAMP;TZID=Asia/Jerusalem:20260105T173000
SUMMARY: CSpecial Event  Faculty Researchers Evening - January 5th | SHE-S Community  at 2026-01-05 17:30:00
DESCRIPTION:We are excited to invite you to an evening of TED talks with pizzas and beers!\nYou are used to meeting them in exercises at the blackboard, but most of their time at the faculty is dedicated to groundbreaking research, now is your chance to hear about their research!Join the TED evening that will include 3 short and fascinating talks by researchers from our faculty.Monday 5.1.26 at 5:30 PM, Graduate Student Lounge, 2nd floor, Taub Building\nEvent schedule:5:30 PM - Gathering, beers and pizzas6:00 PM - Lectures:Noam Koren - PhD student, supervised by Dr. Kira Radinsky. Researcher of multi-participant time series predictionOmer Yizhaq&nbsp;- Master's degree graduate under the supervision of Dr. Shaull Almagor. Researcher of automata jumping over infinite wordsMayan Rivlin - Master's degree student in Prof. Zohar Yakhini's research group. Researcher of optimization and applications in DNA information storage technology and off-target behavior in CRISPRThe event is moderated by Dana Arad - PhD student under the supervision of Dr. Yonatan Belinkov. Researcher of internal mechanisms of language-image models\nRegistration: hereThe event is open to all students in the faculty and no prior knowledge in the research fields is required.\nLooking forward to meeting you!\nSHE S Community
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Graduate Student Lounge, 2nd floor, Taub Building
UID:eventx6a5a287eedc8110916
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20260105T183000
DTEND;TZID=Asia/Jerusalem:20260105T203000
DTSTAMP;TZID=Asia/Jerusalem:20260105T183000
SUMMARY: CSpecial Event  Come be part of the faculty's Capture The Flag - CTF group!! 5.1.26  at 2026-01-05 18:30:00
DESCRIPTION:Come be part of the faculty's Capture The Flag - CTF group!!The meeting will take place on Monday, January 5th at 6:30 PM, at Taub 9.\nWhat's on the meeting?Guest lecture by Uri Bear - From a hacker's perspective on smart and connected vehicles, revealing how automotive systems are hacked and how they can be protected.\nConnected cars are the future, it just makes sense, doesn't it? But as cars connect, there is the increased potential for security risks.\nAutomation, AI, Machine learning and plain old style ECU's contain an ever increasing computation load, an incredibly expanding code base, old and new sensors and algorithms - How do hackers approach all of these?\nA person skilled in reverse engineering and armed with certain tools may be able to eavesdrop on automotive control data. Even more, an advanced hacker could interfere, interact, and modify both the ECU itself and the data flowing across its wires.\nThe cybersecurity landscape is rapidly evolving and the ecosystems are continuously innovating to advance security for devices of all types. The automotive industry is being driven towards a quest for a higher level of security, due to the current plethora of applications, media files, and user inputs available in its systems.\nIn this presentation, I will present:&bull; &ensp;An introduction to automotive computing environment from a hacker's point of view.&bull;&ensp;&ensp;Why is secure hardware a must-have?&bull;&ensp;&ensp;Case study: Hacking a car, near or far.&bull;&ensp;&ensp;Hacking hardware.&bull;&ensp;&ensp;Hacking software.\nMany solutions exist, many are offered, hack yourself to know which are good enough for you.\nThis presentation material was presented at SAE's autonomous vehicle conference in Jan/19 in Israel. None of this material was published.\nShort bio: Uri Bear is an Offensive Security Researcher &amp; Security Evaluation group leader in Intel's Confidential Compute engineering team. Uri has a background in semiconductor design, failure analysis and general hardware and software mischief. Uri has been specializing in security-related forward and reverse engineering for over a decade. Uri holds a M.Sc. degree in Electronic Engineering.\nSponsored by: Intel
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 9
UID:eventx6a5a287eedc9810900
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20260106T143000
DTEND;TZID=Asia/Jerusalem:20260106T153000
DTSTAMP;TZID=Asia/Jerusalem:20260106T143000
SUMMARY: colloq  talk by Micha Sharir (TAU)  Computational Geometry: Per Aspera Ad Astra  at 2026-01-06 14:30:00
DESCRIPTION:I will present an overview of some topics in computational (and combinatorial, and a bit algebraic) geometry, that constitute milestones in the work in this area by myself and by many colleagues and (former) students in the past 45 years. The topics include, as time permits, algorithmic motion planning, arrangements, lower envelopes, incidences, space decomposition, polynomial partitioning, and more.&nbsp;\nBio: Micha Sharir received his Ph.D. in Mathematics from Tel Aviv University in 1976, and then switched to Computer Science, doing his postdoctoral studies at the Courant Institute of New York University. He returned to Tel Aviv University in 1980, and has been there, at the School of Computer Science, ever since. He is also a visiting research professor at the Courant Institute, where he has been the deputy head of the Robotics Lab (1985-89). He has served as the head of the Computer Science Department (twice) and as the head of the School of Mathematics (1997-99). He is one of the co-founders of the Minerva Center for Geometry at Tel Aviv University.&nbsp;\nHis research interests are in computational and combinatorial geometry and their applications. He has pioneered (with Jack Schwartz) the study of algorithmic motion planning in robotics during the early 1980s, and has been involved in many fundamental research studies that have helped to shape the fields of computational and combinatorial geometry. Among his major achievements, in addition to his earlier work on robotics, are the study of Davenport-Schinzel sequences and their numerous geometric applications, the study of geometric arrangements and their applications, efficient algorithms in geometric optimization (including the introduction and analysis of generalized linear programming), and the study of combinatorial problems involving point configurations, including, since 2008, the application of algebraic techniques to problems involving incidences, distinct and repeated distances, and other related problems in combinatorial and computational geometry.&nbsp;\nHis work won him several prizes, including a Max-Planck research prize (1992, jointly with Emo Welzl), the Feher Prize (1999), the Mif'al Hapais' Landau Prize (2002), and the EMET Prize (2007). He is the incumbent of the Nizri Chair in computational geometry and robotics, a Fellow of the Association for Computing Machinery (since 1997), has an honorary doctorate degree from the University of Utrecht (1996), and is a member of the Israeli Academy of Sciences and Humanities (2018). He has supervised 27 Ph.D. students, many of which are now at various stages of an academic career, in Israel and abroad.&nbsp;\nTechnion host: Sarah Keren
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 337
UID:eventx6a5a287eedcde10880
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20260107T123000
DTEND;TZID=Asia/Jerusalem:20260107T140000
DTSTAMP;TZID=Asia/Jerusalem:20260107T123000
SUMMARY: CSpecial Event  Graduate Studies Spotlight Day - STARKWARE  at 2026-01-07 12:30:00
DESCRIPTION:STARKWARE Lecture - The lecture will deal with the topic of Proof Systems and focus on bridging the gap between theory and real-world application (led by Dr. Gil Ben Shachar and Stav Beno).\nRegistration link https://forms.gle/cFiaMjp93DeHtnhT6&nbsp;
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Graduate Student Lounge, 2nd floor
UID:eventx6a5a287eedcfb10913
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20260107T130000
DTEND;TZID=Asia/Jerusalem:20260107T140000
DTSTAMP;TZID=Asia/Jerusalem:20260107T130000
SUMMARY: colloq  talk by Nataly Brukhim  Modern Challenges in Learning Theory  at 2026-01-07 13:00:00
DESCRIPTION:Machine learning relies on its ability to generalize from limited data, yet a principled theoretical understanding of generalization remains incomplete. While binary classification is well understood in the classical PAC framework, even its natural extension to multiclass learning is substantially more challenging.\nIn this talk, I will present recent progress in multiclass learning that characterizes when generalization is possible and how much data is required, resolving a long-standing open problem on extending the Vapnik&ndash;Chervonenkis (VC) dimension beyond the binary setting. I will then turn to complementary results on efficient learning via boosting. &nbsp;We extend boosting theory to multiclass classification, while maintaining computational and statistical efficiency even for unbounded label spaces.\nLastly, I will discuss generalization in sequential learning settings, where a learner interacts with an environment over time. We introduce a new framework that subsumes classically studied settings (bandits and statistical queries) together with a combinatorial parameter that bounds the number of interactions required for learning.\nBio: Nataly Brukhim is a postdoctoral researcher at the Institute for Advanced Study (IAS) and the Center for Discrete Mathematics and Theoretical Computer Science (DIMACS). She received her Ph.D. in Computer Science from Princeton University, where she was advised by Elad Hazan, and was a student researcher at Google AI Princeton. She earned her M.Sc. and B.Sc. in Computer Science from Tel Aviv University.&nbsp;
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 337
UID:eventx6a5a287eedd0d10914
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20260107T143000
DTEND;TZID=Asia/Jerusalem:20260107T153000
DTSTAMP;TZID=Asia/Jerusalem:20260107T143000
SUMMARY: PHD  talk by Aviram Imber  Inferring Election Outcomes under Incomplete Information  at 2026-01-07 14:30:00
DESCRIPTION:Elections are a fundamental mechanism for collective decision-making. Yet in many real-world settings, the available information is often incomplete, inconsistent, or uncertain. This thesis addresses this challenge by developing computational frameworks for analyzing election outcomes under various forms of uncertainty. Our work builds upon the established paradigm of possible and necessary winners, where the goal is to determine which candidates could or must win given partially ordered voter preferences. We advance and broaden this line of inquiry by exploring novel models of incomplete information and new forms of uncertainty across different voting contexts.\nFirst, we generalize the classic possible/necessary winner problem to the problem of computing the minimal and maximal ranks a candidate can possibly obtain. We show that these problems are fundamentally harder than the standard possible/necessary winner problem. Next, we extend the analysis to multi-winner committee selection, proposing new models for incomplete information tailored to approval-based voting. Within these models, we analyze the computational complexity of determining possible and necessary committees. In the third part, we adapt the possible/necessary winner framework to spatial voting, where uncertainty arises from partial information about the positions of voters and candidates in a multidimensional ideological space, rather than from partial preference orders. Within this model we reveal new tractability results. Finally, we move from possibilistic to probabilistic uncertainty. In this setting, voter preferences are fully known, but voter participation is stochastic. We analyze the complexity of computing and approximating the probability of a candidate&rsquo;s victory.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 201
UID:eventx6a5a287eedd2210908
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20260112T103000
DTEND;TZID=Asia/Jerusalem:20260112T113000
DTSTAMP;TZID=Asia/Jerusalem:20260112T103000
SUMMARY: colloq  talk by Idan Attias  Computational and Statistical Limits in Modern Machine Learning  at 2026-01-12 10:30:00
DESCRIPTION:Modern machine learning systems operate in regimes that challenge classical learning-theoretic assumptions. Models are highly overparameterized, trained with simple optimization algorithms, and rely critically on how data is collected and curated. Understanding the limits of learning in these settings requires revisiting both the computational and statistical foundations of learning theory.\nA central question in learning theory asks which functions are tractably learnable. Classical complexity results suggest strong computational barriers, motivating a focus on &ldquo;learnable subclasses&rdquo; defined by properties of the target function. In this talk, I argue for a different perspective by emphasizing the role of the training distribution. Fixing the learning algorithm (e.g. stochastic gradient descent applied to neural networks), I show that allowing a &ldquo;positive distribution shift&rdquo;, where training data is drawn from a carefully chosen auxiliary distribution while evaluation remains on the target distribution, can render several classically hard learning problems tractable.\nBeyond computational considerations, I then study statistical limits of learning in modern, overparameterized models using stochastic convex optimization as a theoretical framework. While classical theory often suggests that successful generalization requires avoiding memorization, I show that memorization is in fact unavoidable: achieving high accuracy requires retaining nontrivial information about the training data and can even enable the identification of individual training examples. These results reveal fundamental privacy&ndash;accuracy tradeoffs inherent to accurate learning.\nBio:&nbsp;Idan Attias is a postdoctoral researcher at the Institute for Data, Econometrics, Algorithms, and Learning (IDEAL), working with Lev Reyzin (University of Illinois Chicago), Nati Srebro, and Avrim Blum (Toyota Technological Institute at Chicago). He obtained his Ph.D. in Computer Science under the supervision of Aryeh Kontorovich (Ben-Gurion University) and Yishay Mansour (Tel Aviv University and Google Research).\nHis research focuses on the foundations of machine learning theory and data-driven sequential decision-making. His work has been recognized with a Best Paper Award at ICML &rsquo;24 and selection as a Rising Star in Data Science (University of California San Diego &rsquo;24). His postdoctoral research is supported by an NSF fellowship, and his Ph.D. studies were fully supported by the Israeli Council for Higher Education Scholarship for Outstanding PhD Students in Data Science.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Auditorium 012, floor 0
UID:eventx6a5a287eedd3710919
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20260112T123000
DTEND;TZID=Asia/Jerusalem:20260112T133000
DTSTAMP;TZID=Asia/Jerusalem:20260112T123000
SUMMARY: ceClub  talk by Tal Zussman (Columbia University)  CE-Club: cache_ext: Customizing the Page Cache with eBPF  at 2026-01-12 12:30:00
DESCRIPTION:The OS page cache is central to the performance of many applications, by reducing excessive accesses to storage. However, its one-size-fits-all eviction policy performs poorly in many workloads. While the systems community has experimented with new and adaptive eviction policies in non-OS settings (e.g., key-value stores, CDNs), it is very difficult to implement such policies in the kernel. To address these shortcomings, we design a flexible eBPF-based framework for the Linux page cache, called cache_ext, that allows developers to customize the page cache without modifying the kernel. cache_ext enables applications to customize the page cache policy for their specific needs, while also ensuring that different applications&rsquo; policies do not interfere with each other and preserving the page cache&rsquo;s ability to share memory across different processes. We demonstrate the flexibility of cache_ext&rsquo;s interface by using it to implement eight different policies, including sophisticated eviction algorithms. Our evaluation shows that it is indeed beneficial for applications to customize the page cache to match their workloads&rsquo; unique properties, and that they can achieve up to 70% higher throughput and 58% lower tail latency.\nBio: Tal Zussman is a PhD student in Computer Science at Columbia University, advised by Prof. Asaf Cidon. He works on operating systems and eBPF, with a focus on accelerating, customizing, and modernizing memory management and storage systems. He received his MS and BS degrees at Columbia, and is an NSF Graduate Research Fellow.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zisapel building 506&nbsp;
UID:eventx6a5a287eedd5010927
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20260112T133000
DTEND;TZID=Asia/Jerusalem:20260112T143000
DTSTAMP;TZID=Asia/Jerusalem:20260112T133000
SUMMARY: TDC Seminar  talk by Liam Brinker (Technion)  TDC Seminar: Self-Stabilizing Algorithms in the Uniform Port Model  at 2026-01-12 13:30:00
DESCRIPTION:We introduce a distributed computational model referred to as the \emph{uniform port} model.An algorithm operating in this model is defined by means of local automata associated with the ports (a.k.a. half-edges) of the input graph.The crux of the uniform port model is that the local automata are identical and admit a constant size description, making the model \emph{truly uniform}.Moreover, since the new model explicitly supports the assignment of (input and) output labels to the graph's (half-)edges, it is much more expressive than existing fully uniform distributed computational models that are restricted to assignments of output labels to the graph's nodes.\nThe main technical contribution of this work is the design of efficient (i.e., with poly-logarithmic time complexity) self-stabilizing uniform port algorithms for various fundamental local symmetry breaking problems, including maximal matching, maximal independent set, and maximal edge and node $c$-coloring.While efficient self-stabilizing algorithms for local symmetry breaking problems have been extensively studied in stronger computational models, our work is the first to demonstrate the existence of such algorithms in a truly uniform model.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 601
UID:eventx6a5a287eedd6510922
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20260113T103000
DTEND;TZID=Asia/Jerusalem:20260113T112900
DTSTAMP;TZID=Asia/Jerusalem:20260113T103000
SUMMARY: colloq  talk by Rana Shahout   Efficient LLM Systems: From Algorithm Design to Deployment  at 2026-01-13 10:30:00
DESCRIPTION:Large Language Models (LLMs) have transformed what machines can do and how systems are designed to serve them. These models are both computationally and memory demanding, revealing the limits of traditional optimization methods that once sufficed for conventional systems. A central challenge in building LLM systems is improving system metrics while ensuring response quality.\nThis talk presents approaches for reducing latency in LLM systems to support interactive applications, from scheduling algorithm design to deployment. It introduces scheduling frameworks that use lightweight predictions of request behavior to make informed decisions about prioritization and memory management across two core settings: standalone LLM inference and API-augmented LLMs that interact with external tools. Across both settings, prediction-guided scheduling delivers substantial latency reductions while remaining practical for deployment.\nBio: Rana Shahout is a Postdoctoral Fellow at Harvard University, working with Michael Mitzenmacher and Minlan Yu. She received her Ph.D. in Computer Science from the Technion and previously worked as a Senior Software Engineer at Mellanox (now NVIDIA). Her research combines machine learning, systems, and algorithmic theory to design efficient and scalable AI systems. Rana is a recipient of the Eric and Wendy Schmidt Postdoctoral Award, the Zuckerman Postdoctoral Fellowship, the Weizmann Institute Women&rsquo;s Postdoctoral Career Development Award, the VATAT Postdoctoral Fellowship, and first place in the ACC Feder Family Award for Best Student Work in Communications.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Auditorium 012, Floor 0
UID:eventx6a5a287eedd7710920
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20260113T113000
DTEND;TZID=Asia/Jerusalem:20260113T123000
DTSTAMP;TZID=Asia/Jerusalem:20260113T113000
SUMMARY: pixel-club  talk by Dr. Itai Lang (The University of Chicago)  Pixel Club: Intuitive and Controllable AI for 3D Geometry  at 2026-01-13 11:30:00
DESCRIPTION:3D geometry powers design, perception, and simulation across various domains, including healthcare, transportation, manufacturing, and entertainment. Yet, traditional workflows for creating and editing 3D content are largely inaccessible, requiring expert knowledge and extensive manual effort. Recent advances in deep learning have enabled the generation of 3D assets through simple interfaces, such as natural language prompts. However, these AI models provide limited control, making it hard to refine the asset iteratively, edit it locally, or regulate the modification strength. As a result, the transformative potential of AI for 3D geometry remains underrealized.\nMy work aims to bridge this gap by pioneering learning-based techniques that combine intuitive user interaction with controllable manipulation and synthesis of 3D geometry. I will show how simple yet expressive interfaces, such as point clicks and text descriptions, can drive core geometry processing and generation tasks, including interactive segmentation, localized texturing, and adjustable surface deformation. These problems pose unique challenges due to the scarcity of high-quality and diverse 3D data. I will show how to address these challenges by designing controllable 3D representations and leveraging the vast prior knowledge encompassed in powerful pretrained AI models for images and language.\nLooking ahead, I will outline my future research agenda, expanding the principles of intuitive and controllable AI to dynamic 3D content that interacts with its environment, and discuss how this direction can advance 3D technologies across different fields.Itai Lang is a postdoctoral researcher at the University of Chicago, working with Professor Rana Hanocka. He received his PhD from Tel Aviv University, where he was advised by Professor Shai Avidan. His research focuses on artificial intelligence for geometry processing, spanning both generative and discriminative tasks. He is particularly interested in developing innovative solutions to make 3D content creation and manipulation intuitive and controllable.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:506, Zisapel Building &amp; Zoom
UID:eventx6a5a287eede7f10923
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20260113T173000
DTEND;TZID=Asia/Jerusalem:20260113T193000
DTSTAMP;TZID=Asia/Jerusalem:20260113T173000
SUMMARY: CSpecial Event  Come Be Part of the Technion's Capture The Flag (CTF) Group!! 12.1.26 | 13.1.26  at 2026-01-13 17:30:00
DESCRIPTION:The meeting will take place on Monday, January 12 at 6:30 PM, in Taub 9 for beginners and Taub 8 for advanced.\nAnd another meeting due to demand on Tuesday, January 13 at 5:30 PM in Taub 3 for advanced and Taub 7 for advanced\nEveryone is welcome, even without previous experience - we look forward to seeing you!\nWhat will we do?\nMonday: Digital forensics - a field where you learn to analyze files, images, memory and networks to uncover hidden information, understand how attacks occurred, and think like a real investigator!\nTuesday: Reverse engineeringYou will learn to dismantle defense mechanisms and think creatively. Additionally, if you are a student interested in taking the course next semester, you can get a small taste of\nThere will be pizzas
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:12.1.26 - Taub 8 + Taub 9 | 13.1.26 - Taub 3 + Taub 7
UID:eventx6a5a287eede9e10925
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20260113T200000
DTEND;TZID=Asia/Jerusalem:20260113T220000
DTSTAMP;TZID=Asia/Jerusalem:20260113T200000
SUMMARY: CSpecial Event  talk by Dr. Nir Rosenfeld   "Technion on the Bar" January - This Time: Machine Learning in a Human World  at 2026-01-13 20:00:00
DESCRIPTION:"Technion on the Bar" January - this time: machine learning in a human world\nLearning systems are already everywhere and affect our daily lives, but are they really adapted to people?\nIn a fascinating lecture, Dr. Nir Rosenfeld will take us through the challenges, the promises, and also the first attempts (more or less successful) to develop learning systems that understand humans - and not just data.\n13.1.26 | Technion on the Bar - January LectureFree admission - with advance registrationReserve a place nowRegister at the link
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:The Black Sheep Pub
UID:eventx6a5a287eedeb410917
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20260114T003000
DTEND;TZID=Asia/Jerusalem:20260114T013000
DTSTAMP;TZID=Asia/Jerusalem:20260114T003000
SUMMARY: CSpecial Event  Amazon is Coming to Meet You at the Technion!  at 2026-01-14 00:30:00
DESCRIPTION:Amazon is coming to meet you at the Technion!\nWednesday, January 14th at 12:30 PM in the Piano Auditorium at Taub, Faculty of Computer Science.\nOn the program (pre-registration required):\n12:30 PM Mingling with the development teams | Taub Lobby\n12:45 PM Technology lectures on the hottest products and trends in the industry, explanation of the Amazon recruitment process and open positions. | Piano Auditorium, Taub\nCome get inspired, ask questions and discover the open positions for students and graduates!\nTo register - click the link
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Auditorium 012, Floor 0
UID:eventx6a5a287eedec910926
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20260114T103000
DTEND;TZID=Asia/Jerusalem:20260114T113000
DTSTAMP;TZID=Asia/Jerusalem:20260114T103000
SUMMARY: ceClub  talk by Rana Shahout (Harvard University)  CE-Club: Efficient LLM Systems: From Algorithm Design to Deployment  at 2026-01-14 10:30:00
DESCRIPTION:Large Language Models (LLMs) have transformed what machines can do and how systems are designed to serve them. These models are both computationally and memory demanding, revealing the limits of traditional optimization methods that once sufficed for conventional systems. A central challenge in building LLM systems is improving system metrics while ensuring response quality.This talk presents approaches for reducing latency in LLM systems to support interactive applications, from scheduling algorithm design to deployment. It introduces scheduling frameworks that use lightweight predictions of request behavior to make informed decisions about prioritization and memory management across two core settings: standalone LLM inference and API-augmented LLMs that interact with external tools. Across both settings, prediction-guided scheduling delivers substantial latency reductions while remaining practical for deployment.\nBio: Rana Shahout is a Postdoctoral Fellow at Harvard University, working with Michael Mitzenmacher and Minlan Yu. She received her Ph.D. in Computer Science from the Technion and previously worked as a Senior Software Engineer at Mellanox (now NVIDIA). Her research combines machine learning, systems, and algorithmic theory to design efficient and scalable AI systems. Rana is a recipient of the Eric and Wendy Schmidt Postdoctoral Award, the Zuckerman Postdoctoral Fellowship, the Weizmann Institute Women&rsquo;s Postdoctoral Career Development Award, the VATAT Postdoctoral Fellowship, and first place in the ACC Feder Family Award for Best Student Work in Communications.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Mayer 861 &amp; Zoom
UID:eventx6a5a287eededd10924
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20260114T130000
DTEND;TZID=Asia/Jerusalem:20260114T140000
DTSTAMP;TZID=Asia/Jerusalem:20260114T130000
SUMMARY: colloq  talk by Noam Razin  Fundamentals of Aligning General-Purpose AI  at 2026-01-14 13:00:00
DESCRIPTION:The field of artificial intelligence (AI) is undergoing a paradigm shift, moving from neural networks trained for narrowly defined tasks (e.g., image classification and machine translation) to general-purpose models such as ChatGPT. These models are trained at unprecedented scales to perform a wide range of tasks, from providing travel recommendations to solving Olympiad-level math problems. As they are increasingly adopted in society, a central challenge is to ensure the alignment of general-purpose models with human preferences. In this talk, I will present a series of works that reveal fundamental pitfalls in existing alignment methods. In particular, I will show that they can: (1) suffer from a flat objective landscape that hinders optimization, and (2) fail to reliably increase the likelihood of generating preferred outputs, sometimes even causing the model to generate outputs with an opposite meaning. Beyond characterizing these pitfalls, our theory provides quantitative measures for identifying when they occur, suggests preventative guidelines, and has led to the development of new data selection and alignment algorithms, validated at large scale in real-world settings. Our contributions address both efficiency challenges and safety risks that may arise in the alignment process. I will conclude with an outlook on future directions, toward building a practical theory in the age of general-purpose AI.\nBio: Noam Razin is a Postdoctoral Fellow at Princeton Language and Intelligence, Princeton University. His research focuses on the fundamentals of artificial intelligence (AI). By combining mathematical analyses with systematic experimentation, he aims to develop theories that shed light on how modern AI works, identify potential failures, and yield principled methods for improving efficiency, reliability, and performance.\nNoam earned his PhD in Computer Science at Tel Aviv University, where he was advised by Nadav Cohen. Prior to that, he obtained a BSc in Computer Science (summa cum laude) at The Hebrew University of Jerusalem under the Amirim honors program. For his research, Noam received several honors and awards, including the Zuckerman Postdoctoral Scholarship, the Israeli Council for Higher Education (VATAT) Postdoctoral Scholarship, the Apple Scholars in AI/ML PhD fellowship, the Tel Aviv University Center for AI and Data Science excellence fellowship, and the Deutsch Prize for PhD candidates.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 401
UID:eventx6a5a287eedeee10921
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20260118T110000
DTEND;TZID=Asia/Jerusalem:20260118T120000
DTSTAMP;TZID=Asia/Jerusalem:20260118T110000
SUMMARY: MSC  talk by Tomer Borreda  Linear Complexity Graph Transformers with Adaptive Hub-Spoke Reassignment  at 2026-01-18 11:00:00
DESCRIPTION:We present ReHub, a novel graph transformer architecture that achieves linear complexity through an efficient reassignment technique between nodes and virtual nodes. Graph transformers have become increasingly important in graph learning for their ability to utilize long-range node communication explicitly, addressing limitations such as oversmoothing and oversquashing found in message-passing graph networks.&nbsp;However, their dense attention mechanism scales quadratically with the number of nodes, limiting their applicability to large-scale graphs. ReHub draws inspiration from the airline industry's hub-and-spoke model, where flights are &nbsp;assigned to optimize operational efficiency.\nIn our approach, graph nodes (spokes) are dynamically reassigned to a fixed number of virtual nodes (hubs) at each model layer. Recent work, Neural Atoms, has demonstrated impressive and consistent improvements over GNN baselines by utilizing such virtual nodes; their findings suggest that the number of hubs strongly influences performance. However, increasing the number of hubs typically raises complexity, requiring a trade-off to maintain linear complexity.\nOur key insight is that each node only needs to interact with a small subset of hubs to achieve linear complexity, even when the total number of hubs is large. To leverage all hubs without incurring additional computational costs, we propose a simple yet effective adaptive reassignment technique based on hub-hub similarity scores, eliminating the need for expensive node-hub computations.\nOur experiments on long-range graph benchmarks indicate a consistent improvement in results over the base method, Neural Atoms, while maintaining a linear complexity instead of O(n^3/2). Remarkably, our sparse model achieves performance on par with its non-sparse counterpart. Furthermore, ReHub outperforms competitive baselines and consistently ranks among the top performers across various benchmarks.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 401
UID:eventx6a5a287eedf0210928
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20260119T130000
DTEND;TZID=Asia/Jerusalem:20260119T140000
DTSTAMP;TZID=Asia/Jerusalem:20260119T130000
SUMMARY: colloq  talk by Jonathan Shafer (MIT)  From Learning Theory to Cryptography: Provable Guarantees for AI  at 2026-01-19 13:00:00
DESCRIPTION:Ensuring that AI systems behave as intended is a central challenge in contemporary AI. This talk offers an exposition of provable mathematical guarantees for learning and security in AI systems.\nStarting with a classic learning-theoretic perspective on generalization guarantees, we present two results quantifying the amount of training data that is provably necessary and sufficient for learning: (1) In online learning, we show that access to unlabeled data can reduce the number of prediction mistakes quadratically, but no more than quadratically [NeurIPS23, NeurIPS25 Best Paper Runner-Up]. (2) In statistical learning, we discuss how much labeled data is actually necessary for learning&mdash;resolving a long-standing gap left open by the celebrated VC theorem [COLT23].\nProvable guarantees are especially valuable in settings that require security in the face of malicious adversaries. The main part of the talk adopts a cryptographic perspective,&nbsp; showing how to: (1) Utilize interactive proof systems to delegate data collection and AI training tasks to an untrusted party [ITCS21, COLT23, NeurIPS25]. (2) Leverage random self-reducibility to provably remove backdoors from AI models, even when those backdoors are themselves provably undetectable [STOC25].\nBio: Jonathan Shafer is a Postdoctoral Associate at MIT, working with Vinod Vaikuntanthan. He co-organizes the MIT ML+Crypto Seminar. Previously, he earned a PhD from UC Berkeley advised by Shafi Goldwasser.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Auditorium 012, Floor 0
UID:eventx6a5a287eedf1910929
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20260119T183000
DTEND;TZID=Asia/Jerusalem:20260119T193000
DTSTAMP;TZID=Asia/Jerusalem:20260119T183000
SUMMARY: CSpecial Event  Come be Bart of the Technion's Capture The Flag (CTF) Group!! 19.1.26  at 2026-01-19 18:30:00
DESCRIPTION:The meeting will take place on Monday, January 19 at 6:30 PM, in Taub 7 for beginners and Taub 8 for advanced\nEveryone is welcome, even without previous experience &ndash; we look forward to seeing you!\nWhat will we do?\nWant to really challenge yourself? The CTF competition is the most fun way to learn cyber: solve real puzzles, think like hackers and gain experience that feels like the real world. No lectures, no digging &ndash; just challenge, action and fast learning suitable for both beginners and advanced. Join us?\nThere will be pizza
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 7 | Taub 8
UID:eventx6a5a287eedf2b10931
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20260120T113000
DTEND;TZID=Asia/Jerusalem:20260120T123000
DTSTAMP;TZID=Asia/Jerusalem:20260120T113000
SUMMARY: MSC  talk by Tom Agami  Automatic Saffron Picking  at 2026-01-20 11:30:00
DESCRIPTION:We introduce a vision-guided robotic system for automated saffron flower harvesting. Using camera-based perception and robotic manipulation, the system detects and cuts whole flowers while preserving their stigmas and avoiding plant damage.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 401
UID:eventx6a5a287eedf3d10876
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20260121T130000
DTEND;TZID=Asia/Jerusalem:20260121T140000
DTSTAMP;TZID=Asia/Jerusalem:20260121T130000
SUMMARY: Theory Semina  talk by Assaf Reiner (Hebrew University)  Theory Seminar: Explicit Lossless Vertex Expanders  at 2026-01-21 13:00:00
DESCRIPTION:Lossless vertex expanders are d-regular (or biregular) graphs in which every small set of vertices S has almost the largest possible number of neighbors d|S|. While random regular graphs are known to be lossless expanders with high probability, constructing them explicitly has been a longstanding challenge.&nbsp;\nIn this talk, I will present the first explicit construction of constant-degree two-sided lossless vertex expanders. The resulting graphs also admit a free group action, and hence realize the new families of good quantum LDPC codes due to Lin and Hsieh. The construction is based on taking an appropriate product of a constant-sized lossless expander with a base graph constructed from LPS Ramanujan cubical complexes, which are natural high-dimensional generalizations of the well-known LPS Ramanujan graphs.\nBased on joint work with Jun-Ting Hsieh, Alex Lubotzky, Sidhanth Mohanty and Rachel Zhang (FOCS 2025).
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 401
UID:eventx6a5a287eedf4d10934
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20260121T150000
DTEND;TZID=Asia/Jerusalem:20260121T160000
DTSTAMP;TZID=Asia/Jerusalem:20260121T150000
SUMMARY: pixel-club  talk by Yigal Koifman  Pixel Club: How Can Hundreds of Simple Agents Coordinate Without Explicit Communication, GPS, or Centralized Control?  at 2026-01-21 15:00:00
DESCRIPTION:Graduate seminar on Distributed and Decentralized Task Allocation in Flexible Swarms by Yigal Koifman, supervised by Prof. Alfred M. Bruckstein and Dr. Ariel Barel.\nMulti-agent systems provide a powerful platform for solving complex tasks in dynamic environments, yet achieving coordination without centralized control or direct communication remains a core challenge. This seminar will present a unified, scalable, and robust framework for distributed task allocation that emerges solely from local interactions in both homogeneous and heterogeneous swarms.\nWe introduce and analyze decentralized control methods based on geometric interaction rules, a flexible swarm model that preserves cohesion and enables adaptive motion, and learning-based algorithms using both neural networks (NN) and reinforcement learning (RL). We further show that swarm steering can be achieved through broadcast cueing without direct communication.\nThe discussion will highlight different methods, algorithmic design, and practical implications for autonomous robotics, sensor networks, and other real-world multi-agent systems.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 301
UID:eventx6a5a287eedf5c10933
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20260121T160000
DTEND;TZID=Asia/Jerusalem:20260121T170000
DTSTAMP;TZID=Asia/Jerusalem:20260121T160000
SUMMARY: ceClub  talk by Eitan Eliav (Technion)  CE-Club: ToggleCCI: Dynamic Cost Optimization for Multi-Cloud Transfers  at 2026-01-21 16:00:00
DESCRIPTION:Cloud computing is now extremely common, and many organizations operate workloads across multiple providers. As a result, cross-cloud data transfer has become a critical requirement in modern distributed systems. New services such as Google Cross-Cloud Interconnect (CCI) offer dedicated, high-throughput links with low per-GB transfer costs. Yet these benefits come with high fixed leasing fees and a provisioning delay of several days, creating a difficult cost-performance trade-off under unknown traffic patterns.\nIn this talk, we study the problem of online cross-cloud routing, where decisions must be made without future knowledge. We show that no online algorithm can guarantee constant-factor optimality. To address this, we introduce ToggleCCI, a simple online algorithm that dynamically switches between VPN and CCI based on recent cost observations while accounting for leasing constraints and delays. Using both synthetic workloads and real-world traffic traces, we demonstrate that ToggleCCI consistently tracks the best static strategy and achieves substantial cost savings across a wide range of scenarios.\nM.Sc. student under the supervision of Prof. Isaac Keslassy.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:&nbsp;Mayer 1061
UID:eventx6a5a287eedf6d10930
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20260125T110000
DTEND;TZID=Asia/Jerusalem:20260125T120000
DTSTAMP;TZID=Asia/Jerusalem:20260125T110000
SUMMARY: MSC  talk by Tom Feldman  Counting d-Dimensional Polycubes, Revisited  at 2026-01-25 11:00:00
DESCRIPTION:\nWe study the efficient enumeration of fixed polycubes in dimensions three and higher, a central problem in enumerative combinatorics with exponential growth. Building on Redelmeier&rsquo;s recursive framework and later refinements that replace a full recursion with combinatorial counting, we introduce an optimization for the terminal levels of the computation tree. The key insight is that the number of valid ways to extend a partial polycube depends only on its local adjacency structure, so the combinatorial counts for the remaining extensions can be updated incrementally as the search proceeds instead of being recomputed at each branch.\n\nWe implemented the resulting algorithm in a parallel and distributed setting, and computed new polycube counts in dimensions three, four, and five, including previously unknown values for three-dimensional polycubes of sizes 23 and 24. Experimental results show consistent speedups of 20&ndash;26 percent over prior methods.\n
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 401
UID:eventx6a5a287eedf7e10932
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20260125T113000
DTEND;TZID=Asia/Jerusalem:20260125T123000
DTSTAMP;TZID=Asia/Jerusalem:20260125T113000
SUMMARY: ceClub  talk by Roi Lipman (FalkorDB)  CE-Club: A Different Approach to Graphs  at 2026-01-25 11:30:00
DESCRIPTION:Guest lecture by Roi Lipman from FalkorDB, which develops a graph database for AI purposes
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 9
UID:eventx6a5a287eedf9110937
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20260126T113100
DTEND;TZID=Asia/Jerusalem:20260126T123100
DTSTAMP;TZID=Asia/Jerusalem:20260126T113100
SUMMARY: MSC  talk by Liran Cohen  Memorization and Unlearning in LLMs Through the Lens of Input Loss Landscapes  at 2026-01-26 11:31:00
DESCRIPTION:Understanding how large language models store, retain, and remove knowledge is critical for interpretability, reliability, and compliance with privacy regulations.My work introduces a geometric perspective on memorization and unlearning by analyzing loss behavior over semantically similar inputs through the Input Loss Landscape.\nI show that retained, forgotten, and unseen examples exhibit distinct patterns that reflect active learning, suppressed knowledge, and ignored information.&nbsp;Building on this observation, I propose REMIND (Residual Memorization In Neighborhood Dynamics), a black-box framework for diagnosing residual memorization. I further introduce a new semantic neighbor generation method that enables controlled exploration of local loss geometry.\nThese contributions provide interpretable insights into knowledge retention and forgetting, and offer practical tools for auditing, debugging, and enhancing transparency in large language models.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 601
UID:eventx6a5a287eedfa010918
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20260127T113000
DTEND;TZID=Asia/Jerusalem:20260127T123000
DTSTAMP;TZID=Asia/Jerusalem:20260127T113000
SUMMARY: pixel-club  talk by Mark Sheinin (Weizmann Institute of Science)  Pixel Club: Reinventing Vision: Artificial Intelligence Through New Artificial Eyes  at 2026-01-27 11:30:00
DESCRIPTION:From minute surface vibrations to fast-occurring events, the world is rich with phenomena humans cannot perceive. Likewise, most computer vision systems are primarily based on &lsquo;conventional&rsquo; cameras, designed to mimic the imaging principles of the human eye, and are therefore equally blind to these ubiquitous phenomena. The talk will cover principles for capturing such hidden phenomena and extracting novel, imperceptible scene information by combining cutting-edge algorithms with emerging imaging technologies. Specifically, we&rsquo;ll discuss leveraging light diffraction for high-speed imaging, painting and tracking heat patterns on textureless objects for structure-from-motion, and capturing tiny surface vibrations to infer scene audio, dynamic object locations, and high-level semantic object properties.&nbsp;Mark Sheinin is an assistant professor at the Computer Science department at the Weizmann Institute of Science. Before that, he was a postdoctoral research associate at Carnegie Mellon University&rsquo;s Robotics Institute. He received his Ph.D. in Electrical Engineering from the Technion&mdash;Israel Institute of Technology in 2019. His work has received the Best Student Paper Award at CVPR 2017 and the Best Paper Honorable Mention Award at CVPR 2022. His research focuses on expanding the capabilities of computer vision beyond conventional imaging.https://www.marksheinin.com/&nbsp;
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:506, Zisapel Building
UID:eventx6a5a287eedfb310936
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20260128T113000
DTEND;TZID=Asia/Jerusalem:20260128T123000
DTSTAMP;TZID=Asia/Jerusalem:20260128T113000
SUMMARY: ceClub  talk by Mahmood Sharif (TAU)  CE-Club: From Attacks to Security-Enhancing Insights in NLP Models  at 2026-01-28 11:30:00
DESCRIPTION:Recent advances in natural language processing (NLP) have given rise to transformative models, including large language models (LLMs) and text retrievers. Still, critical concerns remain regarding the security of these models: chiefly, LLMs can be jailbroken and misused (e.g., to launch cyberattacks), and text retrievers in search applications can be manipulated to prioritize adversary-chosen content. In this talk, I will present our recent efforts toward making LLMs and text retrievers more secure. In particular, I will show how potent attacks can provide explanations for models' vulnerabilities, which, in turn, enable us to enhance security. Crucially, I will also demonstrate how our insights can inform the design of even stronger attacks, establishing a cycle that guides continuous model improvements.\nBased on joint work with Matan Ben-Tov and Mor Geva.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Mayer 1061
UID:eventx6a5a287eedfc510939
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20260128T130000
DTEND;TZID=Asia/Jerusalem:20260128T140000
DTSTAMP;TZID=Asia/Jerusalem:20260128T130000
SUMMARY: colloq  talk by Niv Cohen (New York University)  Addressing the Unexpected - Anomaly Detection and AI Safety  at 2026-01-28 13:00:00
DESCRIPTION:While AI models are becoming an ever-increasing part of our lives, our understanding of their behavior in unexpected situations is drifting even further out of reach. This gap poses significant risks to users, model owners, and society at large.\nIn the first part of the talk, I will overview my research on detecting unexpected phenomena with and within deep learning models. Specifically, detecting (i) anomalous samples, (ii) unexpected model behavior, and (iii) unexpected security threats. In the second part of the talk, I will dive into my recent research on a specific type of unexpected security threat: attacks on image watermarks. I will review such attacks and present my recent work toward addressing them. I will conclude with a discussion of future research directions.\nBio:&nbsp;Niv Cohen is a postdoctoral researcher at the school of Computer Science &amp; Engineering at New York University. He completed his PhD at the Hebrew University under the supervision of Yedid Hoshen. His undergrad was in Physics at the Technion as a part of their excellence program. He is also a recipient of the 2024 Blavatnik Prize for Outstanding Israeli Doctoral Students in Computer Science. Niv has been researching core deep learning questions centered around out-of-distribution phenomena: anomaly detection, watermarking generative AI data, and the limits on erasing concepts from deep models.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Auditorium 012, Floor 0
UID:eventx6a5a287eedfd810938
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20260203T130000
DTEND;TZID=Asia/Jerusalem:20260203T140000
DTSTAMP;TZID=Asia/Jerusalem:20260203T130000
SUMMARY: colloq  talk by Omri Shmueli (NTT Research)  Cryptography in the Quantum Age  at 2026-02-03 13:00:00
DESCRIPTION:Quantum information processing is reshaping both the theory and practice of computer science, with cryptography undergoing this transformation particularly intensely. The interface between quantum computation and cryptography spans a broad and fascinating spectrum of questions. At one end are practical challenges: designing classical protocols that run on an average laptop, yet remain secure against adversaries equipped with large-scale quantum computers. On the theoretical side are questions about the relationship between pseudorandom quantum states and black holes, as well as the development of generalized proof systems in which witnesses may be quantum states rather than classical strings. At the other end lies a vision of the future of communication, asking what forms of cryptography are possible when quantum computers are available not only to adversaries, but also to honest parties.\nIn this talk, I will survey this interface with a focus on my research. I will then present a new cryptographic primitive from my work, called one-shot signatures, which enables new capabilities across several domains: it overcomes key impossibilities in decentralized systems, and allows the realization of quantum cryptographic tasks using only classical communication and local quantum computation. I will conclude by outlining central open challenges and a broader vision for the field&rsquo;s future.\nBio: Omri Shmueli is a Postdoctoral Fellow at the Cryptography and Information Security Laboratory at NTT Research in Sunnyvale, California. Previously, he was a Research Fellow at the Simons Institute at UC Berkeley during summer &lsquo;25 and was hosted by Mark Zhandry at NTT Research for one year. Before that, Omri did his PhD at Tel Aviv University under the co-supervision of Nir Bitansky and Zvika Brakerski. Omri&rsquo;s research focuses on quantum computation and cryptography. His work has been recognized with several awards and scholarships, among these are a Best Paper Award at CRYPTO &rsquo;25, a Rothschild Prize &lsquo;24 Excellence Award, a Tel Aviv University Alex Deutsch &rsquo;22 Excellence Award and a Best Student Paper Award at QIP &rsquo;22. His PhD was fully supported by a Clore Foundation Fellowship.\n&nbsp;
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Auditorium 012, Floor 0
UID:eventx6a5a287eedfeb10945
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20260211T103000
DTEND;TZID=Asia/Jerusalem:20260211T113000
DTSTAMP;TZID=Asia/Jerusalem:20260211T103000
SUMMARY: ceClub  talk by Dr. Aviv Yaish (Yale University)  CE-Club: Deconstructing and Rebuilding Trust in Decentralized Economies  at 2026-02-11 10:30:00
DESCRIPTION:Financial systems are becoming increasingly digital and decentralized, demanding a practical fusion of distributed systems security and economic theory. A key enabler of this change, blockchain technology, promises more private and egalitarian economic mechanisms, built by facilitating consensus between pseudonymous actors. However, the theoretical security of these systems may mask significant real-world risks. In this talk, I will present recent advances in bridging this gap between theory and practice. First, I will discuss the resolution of a decade-old puzzle: the lack of observed attacks on major consensus mechanisms. I will then distill the lessons learnt into a holistic approach to designing robust systems and demonstrate its adoption in practice using several lines of work on the economics and security of modern digital markets, tackling problems including denial-of-service resistance in distributed systems and pseudonymous markets where consumers may cheaply create new identities.\nBio: Aviv is a postdoc at Yale University, where he makes and breaks distributed systems by bridging economic theory and practice. His approach is driven by a philosophy of constructive deconstruction: pushing systems to their limits is key to making them robust. Aviv&rsquo;s work has been recognized across several fields: security (CCS Distinguished Paper award), economics (CBER Best Paper award), and industry (three prizes from the Ethereum Foundation and Flashbots). He earned his Ph.D. in Computer Science from the Hebrew University (HUJI), where he was the sole lecturer for large-scale courses and won a teaching award. During his studies, he served as a research consultant at Matter Labs. His honors include the AIANI and Jabotinsky fellowships, and inclusion in HUJI&rsquo;s top 10 CS teaching staff of &lsquo;20, CBER&rsquo;s Top PhD Graduates of &lsquo;23-&lsquo;24, CBER&rsquo;s Rising Stars of &rsquo;25, and HUJI&rsquo;s 40 Under 40 of &lsquo;25 lists.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:506, Zisapel Building &amp; Zoom
UID:eventx6a5a287eee00010948
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20260211T130000
DTEND;TZID=Asia/Jerusalem:20260211T140000
DTSTAMP;TZID=Asia/Jerusalem:20260211T130000
SUMMARY: MSC  talk by Shaked Dery  Computational Identification of IS-Associated Genomic Architectures  at 2026-02-11 13:00:00
DESCRIPTION:Insertion sequence elements (ISs) are short transposable DNA sequences that play a major role in genomic rearrangements in bacteria. Despite their importance and pervasiveness, IS-associated genomic architectures - defined by the spacing, order, and relative orientation of ISs - have not been systematically characterized. Here, we present a computational pipeline for identifying IS-associated architectures, from both assembled genomes and long-read sequencing data at single-molecule resolution. Applying this approach to large-scale microbial genomic datasets, we detected both known amplification structures and more complex architectures, uncovered a previously uncharacterized inverted-flank structure, and observed distributions of structural configurations across IS families and bacterial taxa within the analyzed dataset, suggesting associations between some structures and IS families.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Biology Auditorium
UID:eventx6a5a287eee01310943
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20260212T160000
DTEND;TZID=Asia/Jerusalem:20260212T180000
DTSTAMP;TZID=Asia/Jerusalem:20260212T160000
SUMMARY: CSpecial Event  Technion Exposure Meeting - 12.2 Cinema-City Glilot  at 2026-02-12 16:00:00
DESCRIPTION:Interested in studying Computer Science at the Technion?Join the exposure meeting on Thursday 12.2 starting at 4:00 PM at Cinema-City Glilot\nTo register for the meeting, click on the link
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Cinema-City Glilot
UID:eventx6a5a287eee02510944
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20260215T103000
DTEND;TZID=Asia/Jerusalem:20260215T113000
DTSTAMP;TZID=Asia/Jerusalem:20260215T103000
SUMMARY: colloq  talk by Dr. Aviv Yaish  Deconstructing and Rebuilding Trust in Decentralized Economies  at 2026-02-15 10:30:00
DESCRIPTION:Financial systems are increasingly decentralized, demanding a practical fusion of systems security and economic theory. A key enabler of this change, blockchain technology, promises more open economic mechanisms, built by facilitating consensus between pseudonymous actors. However, the theoretical security of these systems may mask significant real-world risks that arise due to actor incentives. In this talk, I present my contributions to bridging this gap between theory and practice. I start with the resolution of a decade-old puzzle: the lack of observed attacks on major consensus protocols. Then, I present an incentives-aware approach to decentralized systems, and demonstrate its adoption in practice through three lines of work on: (1) collusion-resistant auctions, (2) denial-of-service resilience in transaction processing, and (3) Sybil-proof market mechanisms. I conclude with an agenda for robust and trustworthy decentralized economies.\nBio: Aviv Yaish is a Postdoctoral Associate at Yale University, where he makes and breaks distributed systems by bridging economic theory and practical systems security. His approach is driven by a philosophy of constructive deconstruction: pushing systems to their limits is key to making them robust. His contributions have been recognized by several communities: security (CCS Distinguished Paper), economics (Annual CBER Best Paper), and industry (bounties from the Ethereum Foundation and Flashbots). He is currently a member of the Initiative for Cryptocurrencies &amp; Contracts (IC3), and is a junior external faculty at Vienna's Complexity Science Hub. He earned his Ph.D. in Computer Science from the Hebrew University, where he served as the sole lecturer for two courses and received a teaching excellence award. During his studies, he was also a research consultant at Matter Labs. His honors include the AIANI and Jabotinsky fellowships, and inclusion in CBER's Rising Stars and the Hebrew University's 40 Under 40 lists.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Auditorium 012, Floor 0
UID:eventx6a5a287eee03710947
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20260215T140000
DTEND;TZID=Asia/Jerusalem:20260215T150000
DTSTAMP;TZID=Asia/Jerusalem:20260215T140000
SUMMARY: MSC  talk by Evyatar Ziv  Quantum Algorithms for Symmetric Cryptanalysis  at 2026-02-15 14:00:00
DESCRIPTION:Differential Cryptanalysis is among the most popular types of attacks used to analyze and find vulnerabilities in modern block ciphers. To conduct a differential attack, one must first find a high-probability differential of the block cipher.\nExisting search methods involve either intimate knowledge of the specific cipher that is being attacked, or use of search algorithms whose runtime rises exponentially with the number of rounds over which the search is conducted.\nWe present a quantum-based search method for high probability iterative differentials whose runtime scales linearly in the number of rounds over which the search is conducted and is completely agnostic of the implementation details of the cipher.&nbsp;
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 601
UID:eventx6a5a287eee04910940
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20260223T100000
DTEND;TZID=Asia/Jerusalem:20260223T110000
DTSTAMP;TZID=Asia/Jerusalem:20260223T100000
SUMMARY: MSC  talk by Ivri Hikri  Ambiguous Strategic Classification  at 2026-02-23 10:00:00
DESCRIPTION:\nA common assumption in strategic classification is that the classifier is made public knowledge. However, it remains unclear if, and why, a system would choose to commit to full disclosure. We study a setting in which regulation requires the system to share some, but not all, of the information. This entails a learning task in which the goal is to jointly learn a classifier and the uncertainty surrounding it. Towards this, we adopt from robust mechanism design the notion of ambiguity, which in our setting permits the learner to reveal a set or range of possible classifiers, and choose one to realize. We investigate how ambiguity affects the learning task, propose efficient algorithms for computing best-responses and training, and empirically explore strategic learning and its outcomes in this novel setting and using our approach.\n
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom
UID:eventx6a5a287eee05910949
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20260225T110000
DTEND;TZID=Asia/Jerusalem:20260225T120000
DTSTAMP;TZID=Asia/Jerusalem:20260225T110000
SUMMARY: MSC  talk by Hila Ziv  Accurate Inference of Protocol State Machines from Binary Executables  at 2026-02-25 11:00:00
DESCRIPTION:Communication protocols define how systems exchange information and coordinate actions. In security contexts, understanding these protocols is crucial for tasks such as finding vulnerabilities and analyzing malware. In practice, protocol specifications are often missing, outdated, or incomplete, forcing analysts to reverse engineer protocol behavior. Manual protocol reverse engineering is slow and requires significant expertise, while current automatic approaches are ill-equipped to deal with real protocols, due to state explosion, long execution times, or limited code coverage.\nIn this seminar, we present PALI, an automated system for learning protocols directly from target systems. PALI employs LLM-driven hybrid analysis to create and expand protocol state machines. It then validates the state machines by combining path-level testing with a minimal consistent DFA inference algorithm, based on carefully selected positive and negative examples. This process yields a minimal and reliable state machine consistent with the target system. We evaluated PALI on a wide variety of real and novel protocols, used by benign systems (such as HTTP and SMTP) and malicious ones (the GH0ST malware), achieving accurate state machines while significantly reducing manual analysis effort.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 601 &amp; Zoom
UID:eventx6a5a287eee06910942
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20260226T113000
DTEND;TZID=Asia/Jerusalem:20260226T123000
DTSTAMP;TZID=Asia/Jerusalem:20260226T113000
SUMMARY: MSC  talk by Jonathan Gal  Incremental Dominating Set  at 2026-02-26 11:30:00
DESCRIPTION:Dominating Set is a fundamental problem in graph theory: given a graph, find a minimum-weight subset of vertices such that every vertex is either selected or shares an edge with a selected vertex.\nIn online settings where vertices arrive sequentially, comparing algorithms against an offline optimum with full knowledge of the input leads to extremely strong lower bounds, where even a simple star graph shows that any online algorithm must have competitive ratio &Omega;(n), with n being the number of vertices, matching the trivial strategy of selecting all vertices.\nWe study the incremental dominating set problem, where the optimal algorithm is constrained to the same choices available to online algorithms. This introduces a benchmark that enables a meaningful comparison between algorithms.\nWe present the first results for vertex-weighted graphs and randomized algorithms in this model. For incremental dominating set, we give an O(&Delta;)-competitive deterministic algorithm and an O(log&sup2;&Delta;)-competitive randomized algorithm, where &Delta; is the maximum degree in the graph.\nWe extend these results to the Connected Dominating Set problem, which requires the dominating set to be connected.&nbsp;When the neighborhood of each arriving vertex is known in advance, we can improve the competitive ratio of deterministic algorithms to be polylogarithmic.\nFinally, we establish matching lower bounds, showing that all our results are optimal up to constant factors.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 9 &amp; Zoom
UID:eventx6a5a287eee07c10935
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20260311T113000
DTEND;TZID=Asia/Jerusalem:20260311T123000
DTSTAMP;TZID=Asia/Jerusalem:20260311T113000
SUMMARY: ceClub  talk by Ariel Engelman (Technion)  CE-Club: Seize the Moment: An Information Gain Guided Split for Scaling Neural Network Verification  at 2026-03-11 11:30:00
DESCRIPTION:Verifying the local robustness of neural networks is crucial for understanding their safety level.&nbsp;However, the complexity of a complete analysis is exponential in the number of unstable neurons, which introduce nonlinearity.&nbsp;To scale, many complete verifiers split the verification task into smaller subtasks and select a split by relying on heuristics or learning.\nWe study the problem of finding optimal splits and phrase it as information gain maximization whose goal is to reduce the number of unstable neurons.The challenge is that the information gain is defined over probabilities that are intractable to compute.\nWe present SIGNAL which efficiently computes a split by relying on a differentiable estimate of this information gain.&nbsp;Our key idea is to extend moment propagation to the setting of local robustness verification.\nWe prove that our differentiable estimate converges to the true value as the neurons' input dimensionality grows.&nbsp; We integrate SIGNAL in the alpha-beta-CROWN verifier.&nbsp;For fully-connected networks, whose neurons have high input dimensionality, SIGNAL scales prior approaches by at least 1.6x on average.For the task of computing the largest robust $\epsilon$-ball, within 2 hours, SIGNAL computes 1.25x larger radius for fully-connected and convolutional networks.\nMSc seminar. Supervisor: Dr. Dana Drachsler Cohen
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom
UID:eventx6a5a287eee1f010957
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20260311T130000
DTEND;TZID=Asia/Jerusalem:20260311T140000
DTSTAMP;TZID=Asia/Jerusalem:20260311T130000
SUMMARY: ceClub  talk by Alon Nemirovsky (Technion)  CE-Club: Real-Time Speech Source Separation and Dereverberation in Dynamic Acoustic Environment  at 2026-03-11 13:00:00
DESCRIPTION:Real-time speech source separation in a dynamic environment poses significant challenges for wearable augmented reality (AR) devices due to moving sources, head rotations, and adverse acoustic conditions. This seminar presents a robust bilinear framework that integrates minimum power distortionless response (MPDR) beamforming with weighted prediction error (WPE) dereverberation.\nBy decoupling spatial and temporal filtering, we enable efficient recursive least squares (RLS) adaptation that tracks changes in the acoustic scene. To further improve robustness against steering vector errors caused by direction-of-arrival (DOA) mismatches, we introduce region-of-interest (ROI) beamforming.\nAdditionally, we present a linearly constrained minimum power (LCMP) extension to enable flexible spatial control. Our comprehensive analysis of the framework, combined with robust ROI and LCMP extensions validated on real-world SPEAR recordings, establishes a practical and efficient solution for real-time audio enhancement in wearable AR systems.\nM.Sc. student under the supervision of Prof. Israel Cohen
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom
UID:eventx6a5a287eee21710958
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20260318T140000
DTEND;TZID=Asia/Jerusalem:20260318T150000
DTSTAMP;TZID=Asia/Jerusalem:20260318T140000
SUMMARY: MSC  talk by Ido Amit  Evaluating LLM Uncertainty in Long-Form Generation Using Deterministic Ground Truth  at 2026-03-18 14:00:00
DESCRIPTION:As LLMs generate increasingly long outputs, effective uncertainty estimation must identify errors at fine-grained levels rather than discard entire responses. While such methods exist, evaluating uncertainty at any resolution (token to an entire generation) is challenging and highly sensitive to label imperfections, making zero-noise benchmarks essential; yet, long-form generation benchmarks tend to rely on fallible labels rather than deterministic ground truth.\nWe introduce Single-answer Atomic Long-form Target (SALT), a benchmark of six procedurally generated tasks with single deterministic long textual ground truths, enabling unit-level evaluation of correctness, calibration, and ranking without external judges. Equipped with SALT, our analysis of 50+ LLMs reveals key insights: We identify which confidence functions dominate each uncertainty aspect and show that effective ranking benefits more from coarser evaluation resolutions; SALT further facilitates precise calibration tracking throughout generation, revealing a divergence in the accuracy&ndash;calibration relationship, with high- and low-performing models exhibiting degradation ($\rho=0.87$) and improvement ($\rho=-0.92$).\nFinally, we demonstrate that reasoning, via Chain-of-Thought prompting or internalized through training, introduces a trade-off, improving accuracy while degrading confidence ranking. These findings directly impact risk-critical applications requiring reliable error identification and mitigation.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom
UID:eventx6a5a287eee26910956
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20260319T160000
DTEND;TZID=Asia/Jerusalem:20260319T170000
DTSTAMP;TZID=Asia/Jerusalem:20260319T160000
SUMMARY: MSC  talk by Gal Pomerants  Mechanisms of Repeat Detection in Protein Language Models  at 2026-03-19 16:00:00
DESCRIPTION:Protein sequences are abundant in repeating segments, both as exact copies and as approximate segments with mutations. These repeats are important for protein structure and function, motivating decades of algorithmic work on repeat identification. Recent work has shown that protein language models (PLMs) identify repeats, by examining their behavior in masked-token prediction.\nTo elucidate their internal mechanisms, we investigate how PLMs detect both exact and approximate repeats. We find that the mechanism for approximate repeats functionally subsumes that of exact repeats.\nWe then characterize this mechanism, revealing two main stages: PLMs first build feature representations using both general positional attention heads and biologically specialized components, such as neurons that encode amino-acid similarity. Then, induction heads attend to aligned tokens across repeated segments, promoting the correct answer.\nOur results reveal how PLMs solve this biological task by combining language-based pattern matching with specialized biological knowledge, thereby establishing a basis for studying more complex evolutionary processes in PLMs.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom
UID:eventx6a5a287eee28410955
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20260326T103000
DTEND;TZID=Asia/Jerusalem:20260326T113000
DTSTAMP;TZID=Asia/Jerusalem:20260326T103000
SUMMARY: PHD  talk by Raissa Nataf  Delaying the Future  at 2026-03-26 10:30:00
DESCRIPTION:Distributed asynchronous systems often require explicit synchronization to ensure the correct implementation of shared objects. In this talk, I introduce the Delaying the Future approach for reasoning about the ordering of events in distributed executions. Its key idea is that, under certain conditions, events can be postponed without any process noticing the change.\nI will show how this technique leads to characterizations of communication requirements in asynchronous message-passing systems and in shared-memory systems under the TSO memory model. The Delaying the Future approach provides a unified way to understand the synchronization required by linearizable implementations of common objects such as registers, stacks, and snapshots.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom
UID:eventx6a5a287eee29910951
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20260329T180000
DTEND;TZID=Asia/Jerusalem:20260329T200000
DTSTAMP;TZID=Asia/Jerusalem:20260329T180000
SUMMARY: CSpecial Event  Project Showcase & Information Evening - 29.3.26  at 2026-03-29 18:00:00
DESCRIPTION:Debating which project to choose? Join us for an evening full of innovation, where faculty members and lab directors will present the leading projects for the coming year.\nThis is your opportunity to hear firsthand about the research, ask questions, and find the exact spot for you.\n29.3, Taub 1, 6:00 PM\nRegister at the link
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 1
UID:eventx6a5a287eee2ae10952
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20260330T103000
DTEND;TZID=Asia/Jerusalem:20260330T113000
DTSTAMP;TZID=Asia/Jerusalem:20260330T103000
SUMMARY: PHD  talk by Dean Zadok  Towards Restoring Native Finger Motions from Lower Arm Muscles Behavior for Hand-Traumatized Conditions  at 2026-03-30 10:30:00
DESCRIPTION:Powered prosthetic hands are frequently abandoned due to limited dexterity and unintuitive control. Most commercial devices rely on surface electromyography (sEMG) and support only grasping gestures, falling short of the fine, continuous finger motions required for everyday tasks such as typing on a keyboard or playing a musical instrument.\nIn this talk, we present a series of studies addressing this gap from three complementary angles. First, we introduce an end-to-end system that infers fine finger motions in real time by modeling the hand as a robotic manipulator and encoding muscle dynamics from ultrasound video. Second, we present a low-cost, 3D-printed prosthetic hand engineered for enhanced dexterity, featuring adjustable finger spacing, a two-degree-of-freedom wrist, and independent finger pressing. Third, we propose SonoRank, a step towards calibration-free finger flexion detection from forearm ultrasound.\nSonoRank learns to rank ultrasound sequence pairs by relative motion magnitude, then fine-tunes using a rest reference to classify active flexion across all five fingers without user training data. Together, these papers advance prosthetic control toward practical, calibration-free deployment with fine-finger activation, bringing us closer to restoring native hand function for individuals with upper-limb amputation.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom
UID:eventx6a5a287eee2c010961
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20260407T093000
DTEND;TZID=Asia/Jerusalem:20260407T103000
DTSTAMP;TZID=Asia/Jerusalem:20260407T093000
SUMMARY: PHD  talk by Sally Turutov  Constraint-Aware Machine Learning for Molecular Design  at 2026-04-07 09:30:00
DESCRIPTION:Machine learning has become a central tool in molecular design, enabling the generation and optimization of candidate therapeutics across vast chemical spaces; however, real-world drug development is governed by multiple practical constraints that are rarely addressed jointly in computational models. Beyond optimizing chemical and pharmacological properties, viable molecules must avoid intellectual property conflicts, exhibit tissue-specific biological activity, and remain relevant when translating from preclinical animal models to humans&mdash;constraints that, if ignored, often yield candidates that are computationally promising yet infeasible in practice.\nIn this work, we develop machine learning frameworks that explicitly incorporate these legal, biological, and translational constraints into molecular design and evaluation, demonstrating that constraint-aware approaches produce candidates that are more realistic and better aligned with the requirements of real-world drug discovery.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom&nbsp;
UID:eventx6a5a287eee2d610963
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20260413T123000
DTEND;TZID=Asia/Jerusalem:20260413T133000
DTSTAMP;TZID=Asia/Jerusalem:20260413T123000
SUMMARY: MSC  talk by George Haddad  Approximating the Number of Relevant Variables in a Parity Implies Proper Learning  at 2026-04-13 12:30:00
DESCRIPTION:Consider a model in which we can access a parity function through random uniformly distributed labeled examples in the presence of random classification noise. In this thesis, we study learning in this model and show that approximating the number of relevant variables of a parity function is as hard as properly learning it.\nMore specifically, let $\gamma : \mathbb{R}^+ \to \mathbb{R}^+$ be any strictly increasing function satisfying $\gamma(x) \ge x$. In our first result, we show that from any polynomial-time algorithm that returns a $\gamma$-approximation $D$ (i.e., $\gamma^{-1}(d(f)) \leq D \leq \gamma(d(f))$), of the number of relevant variables~$d(f)$ of any parity function $f$, we can, in polynomial time, construct a solution to the long-standing open problem of polynomial-time learning $k(n)$-sparse parities (parities with $k(n)\le n$ relevant variables), where $k(n) = \omega_n(1)$.\nIn our second result, we show that from any $T(n)$-time algorithm that, for any parity $f$, returns a $\gamma$-approximation of the number of relevant variables $d(f)$ of $f$, we can, in polynomial time, construct a $poly(\Gamma(n))T(\Gamma(n)^2)$-time algorithm that properly learns parities, where $\Gamma(x)=\gamma(\gamma(x))$.\nIf $T(\Gamma(n)^2)=\exp({o(n/\log n)})$, this would resolve another long-standing open problem of properly learning parities in the presence of random classification noise in time~$\exp({o(n/\log n)})$.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 601 &amp; Zoom
UID:eventx6a5a287eee2e810960
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20260415T093000
DTEND;TZID=Asia/Jerusalem:20260415T103000
DTSTAMP;TZID=Asia/Jerusalem:20260415T093000
SUMMARY: MSC  talk by Muhammad Ghoummaid  Contradiction Detection of Clinical Texts  at 2026-04-15 09:30:00
DESCRIPTION:Identifying conflicting claims in biomedical literature is critical for advancing scientific understanding, yet the scarcity of high-quality training data remains a significant challenge. We introduce EvoNLI, an evolutionary algorithm that learns how to transform entailing sentence pairs into challenging contradictions by mutating words until a frozen teacher model confidently flips its prediction, while preserving topical coherence. EvoNLI, applied to PubMed randomized controlled trials (RCTs), generates SciCon&mdash;a dataset of premise&ndash;hypothesis pairs automatically labeled with 94.4% precision, as verified by domain experts. Fine-tuning large language models on SciCon improves contradiction ROC-AUC consistently across eight biomedical NLI benchmarks. EvoNLI and SciCon are publicly available to support contradiction-aware biomedical search and evidence synthesis, and to advance robust domain-specific contradiction detection.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom&nbsp;
UID:eventx6a5a287eee2fd10968
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20260415T123000
DTEND;TZID=Asia/Jerusalem:20260415T133000
DTSTAMP;TZID=Asia/Jerusalem:20260415T123000
SUMMARY: MSC  talk by Arkadi Piven   Modern Digital Pathology  at 2026-04-15 12:30:00
DESCRIPTION:Recent advances in computer vision, foundation models, and transformer architectures have transformed computational pathology, enabling deep learning systems to extract clinically actionable information directly from digitized tissue slides. This seminar explores how these technologies come together in modern digital pathology frameworks, and presents two studies demonstrating their clinical impact.\nThe first study addresses a critical diagnostic gap in low-resource settings, showing that convolutional neural networks applied to Giemsa-stained bone marrow aspirates can predict B/T-cell lineage and ETV6&ndash;RUNX1 translocation status in pediatric acute lymphoblastic leukemia &mdash; tasks that traditionally require expensive molecular assays unavailable in many parts of the world.\nThe second study tackles overtreatment in breast cancer. The TAILORx trial established that adjuvant chemotherapy can be spared for postmenopausal HR+/HER2&minus; node-negative breast cancer patients with a 21-gene Recurrence Score (RS) of 11&ndash;25. However, among premenopausal women with RS 16&ndash;25, a small benefit from chemotherapy could not be ruled out. Consequently, guidelines suggest considering chemotherapy for this population, creating a therapeutic dilemma and leading to widespread overtreatment of patients who may not benefit from chemotherapy. Using deep survival analysis on H&amp;E whole-slide images, we identify which women in this group truly benefit from adjuvant chemotherapy. Our model stratifies 76% of this population as low-risk, for whom chemotherapy can be safely omitted, while correctly identifying the high-risk subset that benefits from treatment.\nTogether, these works illustrate how digital pathology can democratize access to precision diagnostics and enable more personalized, less toxic cancer care.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 401 &amp; Zoom
UID:eventx6a5a287eee31110959
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20260415T150000
DTEND;TZID=Asia/Jerusalem:20260415T160000
DTSTAMP;TZID=Asia/Jerusalem:20260415T150000
SUMMARY: ceClub  talk by Alon Levi (Technion)  CE-Club: Elastic Cache Provisioning for Multithreaded CPUs Using the Multi-Amdahl Approach  at 2026-04-15 15:00:00
DESCRIPTION:Modern multi-threaded processors increasingly experience diminishing performance returns from thread-level parallelism (TLP) due to contention for shared cache resources. While wider cores and increased thread counts are intended to improve throughput, inter-thread cache interference and non-uniform memory access behaviors often negate these benefits, particularly under heterogeneous and dynamically changing workloads. Existing mitigation techniques, including software-managed cache partitioning and architecture-specific control mechanisms, have shown limited scalability and poor portability across platforms.\nThis thesis investigates adaptive cache management as a means to address these limitations and introduces a Multi-Amdahl&ndash;based microarchitectural framework for dynamic cache provisioning. The proposed approach leverages the Multi-Amdahl principle to model how individual thread performance responds to incremental changes in allocated cache capacity, thereby quantifying each thread&rsquo;s performance sensitivity to cache resources. By continuously estimating these sensitivities at runtime, the proposed mechanism dynamically redistributes cache capacity to maximize overall system throughput while preserving fairness among concurrent threads.\nTo enable cache resources partitioning, the thesis presents an elastic cache microarchitecture that supports fine-grained, runtime reallocation of cache blocks among threads. This elasticity allows per-thread cache footprints to expand or contract in response to evolving workload demands, enabling effective cache partitioning under constrained resources. The proposed techniques are evaluated using detailed, simulation-based experiments on the SPEC CPU2017 benchmark suite. Results demonstrate performance improvements of up to 23% in instructions per cycle, reductions of up to 74% in cache miss rates, and gains of up to 29% in weighted speedup compared to baseline cache partitioning schemes.\nOverall, this work demonstrates that Multi-Amdahl&ndash;based modeling provides a practical, scalable foundation for adaptive cache management and offers a viable path toward improving performance and efficiency in future heterogeneous multi-core processor architectures.\nMSc seminar. Supervisor: Prof. Uri Weiser and Prof. Freddy Gabbay.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom
UID:eventx6a5a287eee32910969
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20260415T153000
DTEND;TZID=Asia/Jerusalem:20260415T163000
DTSTAMP;TZID=Asia/Jerusalem:20260415T153000
SUMMARY: MSC  talk by Noam Krupnik  On Spectral Graph Determination and Transitivity of Gilbert Graphs  at 2026-04-15 15:30:00
DESCRIPTION:The study of spectral graph determination is a central and fascinating topic in spectral graph theory and algebraic combinatorics. This area investigates the spectral characterization of various classes of graphs, develops methods for constructing and distinguishing cospectral nonisomorphic graphs, and analyzes the conditions under which the spectrum of a graph uniquely determines its structure. In the first part of the seminar, we present both classical results and recent advances in spectral graph determination.\nThe study of graph symmetries and different notions of transitivity is also of fundamental interest in algebraic graph theory. In the second part of the talk, we examine transitivity properties of Gilbert graphs and their complements, and discuss the main ideas underlying these results.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom
UID:eventx6a5a287eee34310953
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20260416T123000
DTEND;TZID=Asia/Jerusalem:20260416T133000
DTSTAMP;TZID=Asia/Jerusalem:20260416T123000
SUMMARY: MSC  talk by Netanel Arussy  Searching for Underlying Trajectories with Self-Organizing Maps  at 2026-04-16 12:30:00
DESCRIPTION:Self-Organizing Maps are unsupervised machine learning algorithms used primarily for clustering and dimensionality reduction. They map high-dimensional data for improved interpretability using a competitive learning approach. Each data point is mapped with some distance to a point on the map. These distances, called activations, show underlying trajectories in the data that can be explored. This is done in two studies.\nThe first study seeks vulnerabilities in public data by using self-organizing maps to bring people's sensitive attributes to the surface. This can reveal sensitive attributes with low correlation to the data are recoverable, thus leaving people's personal data at risk.\nThe second study looks into finding underlying trajectories in cell types. Cell type trajectory inference, also called pseudotime analysis, maps developmental and state changes in cells. Using trajectory inference to order single-cell omics data is used in stem cell differentiation, disease progression, and cell response to stimuli among other things.\nThese studies open the door to new research into applications of self-organizing maps from a 3rd dimension.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom
UID:eventx6a5a287eee35510962
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20260416T130000
DTEND;TZID=Asia/Jerusalem:20260416T140000
DTSTAMP;TZID=Asia/Jerusalem:20260416T130000
SUMMARY: MSC  talk by Firas Yazbak  Koopman-Based Stochastic Differential Equations for Time Series Forecasting  at 2026-04-16 13:00:00
DESCRIPTION:Koopman-based methods for time-series forecasting model nonlinear dynamics as linear evolution in a latent space, but deterministic formulations often mix noise with the underlying dynamics, limiting long-term accuracy. This talk introduces KoopSDE, which extends Koopman models with a latent stochastic differential equation.\nBy linking the drift to the Koopman generator and learning a diffusion term, the method separates dynamics from noise and improves robustness in long-horizon forecasting.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom
UID:eventx6a5a287eee36810965
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20260416T140000
DTEND;TZID=Asia/Jerusalem:20260416T150000
DTSTAMP;TZID=Asia/Jerusalem:20260416T140000
SUMMARY: MSC  talk by Alon Hacohen  Enabling Single-Cell Foundation Models for Scarce Data via Dependency-Aware Masking  at 2026-04-16 14:00:00
DESCRIPTION:The adaptation of Large Language Model architectures to computational biology has enabled Single-Cell Foundation Models for learning from single-cell RNA-sequencing data. However, many of these models rely on masking strategies from natural language processing; unlike words in a sentence, gene expression is governed by highly correlated regulatory networks, making random masking or structure naive techniques biologically misaligned. Viewed through an information-theoretic lens, this introduces a key inefficiency: models can reconstruct masked genes from local correlations, limiting their ability to learn accurate biological representations based on higher-order structure and driving reliance on large datasets - a challenge in data-scarce settings such as rare disease cohorts or privacy-preserving environments.\nTo address this, we introduce domain-informed masking during pre-training. In this talk, we present CorrMask, a data-driven, dependency-aware masking scheme that leverages gene correlation structure to jointly mask related genes, encouraging learning from global cellular context. Across tissue-specific datasets, CorrMask matches baseline performance on both cell- and gene-level tasks using less data, with the strongest gains in underrepresented cell populations.\nThese results position CorrMask as an effective &ldquo;data multiplier&rdquo; for enabling efficient, biologically grounded foundation models, with broader implications for predictive modeling in our field.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom
UID:eventx6a5a287eee37810967
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20260416T163000
DTEND;TZID=Asia/Jerusalem:20260416T173000
DTSTAMP;TZID=Asia/Jerusalem:20260416T163000
SUMMARY: MSC  talk by Evan Abboud   Tight Algorithm and Hardness for Submodular Linear Ordering  at 2026-04-16 16:30:00
DESCRIPTION:We consider the Minimum Linear Ordering Problem: given a ground set N of cardinality n and a non-negative set function f: 2^N&rarr;R&ge;0, the goal is to find an ordering &pi; of N that minimizes the sum of the values of f over all prefixes of &pi;. This problem has been studied for various classes of set functions, and the case of a submodular f is of special interest, as it captures classic problems including Minimum Linear Arrangement and Minimum Containing Interval Graph. In this work, we resolve the approximability of the Minimum Linear Ordering Problem for a general submodular f by establishing matching upper and lower bounds and present: (1) a polynomial-time algorithm achieving an O(&radic;(n/ln n))-approximation; and (2) a matching information-theoretic hardness result, showing that no algorithm evaluating f a polynomial number of times can achieve an o(&radic;(n/ln n))-approximation. Previously, the best known hardness of approximation was 2, and an O(&radic;(n/ln n))-approximation was known only for the special case where f is both submodular and symmetric.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom
UID:eventx6a5a287eee38b10970
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20260423T140000
DTEND;TZID=Asia/Jerusalem:20260423T150000
DTSTAMP;TZID=Asia/Jerusalem:20260423T140000
SUMMARY: MSC  talk by Dor Noti  DART: Efficient Anytime Task Assistance Planning in Continuous  Spaces  at 2026-04-23 14:00:00
DESCRIPTION:Task Assistance Planning (TAP) involves coordinating an assisting robot to support a task robot executing a predefined trajectory, with the goal of maximizing the duration of effective assistance to the task robot. This coordination may be critical in diverse applications such as providing a communication relay in search-and-rescue missions. While existing approaches optimally solve TAP on static, precomputed discrete roadmaps, they scale poorly to continuous configuration spaces because committing to a fixed roadmap can either omit optimal solutions or incur prohibitive preprocessing delays. Furthermore, a naive anytime extension that interleaves continuous roadmap densification with optimal discrete solvers introduces cyclic temporal dependencies that render the problem computationally intractable. To address this gap, we introduce Directed Acyclic Roadmap for TAP (DART), an anytime algorithmic framework designed for continuous TAP. DART restricts the incrementally constructed roadmap to a Directed Acyclic Graph (DAG). This key architectural shift eliminates cyclic temporal dependencies, drastically reducing the combinatorial search space evaluated during path optimization. We evaluate DART on both low-dimentional simulated real-world problem and high-dimentional synthetic problem and demonstrate empirically that DART achieves computational speedups of up to three orders of magnitude compared to undirected baseline methods, allowing for rapid incremental updates and the discovery of higher-quality assistance trajectories within a given time budget.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom
UID:eventx6a5a287eee39610971
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20260423T170000
DTEND;TZID=Asia/Jerusalem:20260423T180000
DTSTAMP;TZID=Asia/Jerusalem:20260423T170000
SUMMARY: PHD  talk by Dan Kalifa  Enhancing Protein Language Models with Multidimensional Protein Knowledge  at 2026-04-23 17:00:00
DESCRIPTION:Proteins are fundamental to biological systems, and accurately representing them is essential for understanding biological function and drug discovery. Recent protein language models learn representations from amino acid sequences, yet proteins are inherently multidimensional, characterized by structure, dynamics, and molecular interactions. This thesis investigates how integrating multidimensional protein knowledge can enhance protein language models and improve biological understanding.\nWe introduce GOProteinGNN for integrating protein knowledge graphs, FusionProt for sequence&ndash;structure fusion, ProtLigand for leveraging protein&ndash;ligand interactions, and DynamicsPLM for modeling conformational dynamics. Across diverse biological tasks, these approaches improve performance and produce biologically meaningful representations.\nThis research was also applied in a drug discovery laboratory, demonstrating the practical value of multidimensional protein representations. Overall, this thesis shows that integrating functional, structural, dynamic, and interaction-based information substantially enhances protein representation learning and supports advances in biomedical research.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom
UID:eventx6a5a287eee3a210966
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20260426T153000
DTEND;TZID=Asia/Jerusalem:20260426T163000
DTSTAMP;TZID=Asia/Jerusalem:20260426T153000
SUMMARY: MSC  talk by Lyan Abboud  Error-Correcting Codes for the Sum Channel  at 2026-04-26 15:30:00
DESCRIPTION:We introduce the sum channel, a new channel model motivated by applications in distributed storage and DNA data storage. In the error-free case, it takes as input an $\ell$-row binary matrix and outputs an $(\ell+1)$-row matrix whose first $\ell$ rows equal the input and whose last row is their parity (sum) row.\nWe construct a two-deletion-correcting code with redundancy $2\lceil\log_2\log_2 n\rceil+ \log_2 \ell + O(1)$ for $\ell$-row inputs. When $\ell=2$, we establish a lower bound of $\lceil\log_2\log_2 n\rceil + O(1)$ bits, implying that our redundancy is optimal up to a factor of 2.\nWe also present a code correcting a single substitution with $\lceil \log_2(\ell+1)\rceil$ redundant bits and prove that it is within one bit of optimality.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:ZOOM
UID:eventx6a5a287eee3ad10972
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20260429T113000
DTEND;TZID=Asia/Jerusalem:20260429T123000
DTSTAMP;TZID=Asia/Jerusalem:20260429T113000
SUMMARY: ceClub  talk by Prof. Adam Morrison (Tel Aviv University)  CE-Club: Arm Weak Memory Consistency on Apple Silicon: What Is It Good For?  at 2026-04-29 11:30:00
DESCRIPTION:Weak memory models such as the Arm model are perceived as enabling higher performance than strong models such as TSO. We critically test this perception on Apple silicon CPUs, whose runtime-configurable TSO mode enables a direct comparison with native Arm mode. We find that Apple silicon TSO mode preserves Arm weak-memory optimizations, typically yielding execution times within 3% of Arm mode across modern applications and classic benchmarks. Although some applications experience higher TSO slowdowns, we trace these to artifacts of Apple's TSO implementation rather than inherent TSO ordering constraints. Our results challenge the perception that the Arm memory model offers a significant performance advantage over TSO in Apple silicon.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zisapel building 506 &amp; Zoom
UID:eventx6a5a287eee3b810973
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20260504T103000
DTEND;TZID=Asia/Jerusalem:20260504T113000
DTSTAMP;TZID=Asia/Jerusalem:20260504T103000
SUMMARY: colloq  talk by Guy Amir (Cornell University)  A Semantic Approach to Verifying Programmable Networks  at 2026-05-04 10:30:00
DESCRIPTION:As networks become more programmable, they are increasingly built around flexible software components. While this programmability enables new functionality and faster innovation, it also makes network behavior harder to reason about. In this talk, I will present a research agenda that brings ideas from formal methods to programmable networks. In particular, I will present techniques that leverage programmable-network semantics for concurrency safety, traffic monitoring, and failure recovery. More broadly, this work illustrates how semantic foundations can help bring stronger correctness guarantees to modern networked systems.\nBio: Guy Amir is a Postdoctoral Researcher at Cornell University, conducting research at the intersection of formal methods, networking, and systems. He earned his Ph.D. in 2024 from the Hebrew University of Jerusalem, where he studied AI safety, focusing on formally verifying reactive AI systems and interpreting neural networks. He holds an M.Sc. in Computer Science and a B.Sc. in Computational Biology and Computer Science, both from the Hebrew University. He has received Rothschild, Fulbright, AI-Net, and Charles Clore fellowships, as well as an ICML Spotlight and KLA Award.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Auditorium 012, Floor 0
UID:eventx6a5a287eee3c310975
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20260505T113000
DTEND;TZID=Asia/Jerusalem:20260505T123000
DTSTAMP;TZID=Asia/Jerusalem:20260505T113000
SUMMARY: pixel-club  talk by Omri Hirsch (Ben Gurion University)  Pixel Club: FastJAM: a Fast Joint Alignment Model for Images  at 2026-05-05 11:30:00
DESCRIPTION:Joint Alignment (JA) aims to align a collection of images into a shared coordinate frame such that semantically corresponding features coincide spatially. Despite its importance in many vision applications, existing JA methods often rely on heavy optimization pipelines, large-capacity models, and extensive hyperparameter tuning, leading to long training times and limited scalability.This talk presents FastJAM, a fast and lightweight joint alignment framework that reframes JA as a graph-based learning problem over sparse keypoints. FastJAM leverages pairwise correspondences from an off-the-shelf matcher and a graph neural network to efficiently predict per-image homography transformations, achieving state-of-the-art alignment quality while reducing runtime from minutes or hours to just seconds.Link to Paper: https://bgu-cs-vil.github.io/FastJAM/&nbsp;\nOmri Hirsch is an MSc. student in Computer Science at Ben-Gurion University of the Negev, conducting research in Computer Vision and Machine Learning in The Vision, Inference, and Learning (VIL) group under the supervision of Prof. Oren Freifeld. His research focuses on efficient geometric learning and joint image alignment, and he is the first author of FastJAM, recently accepted to NeurIPS 2025. He has previously worked on medical imaging and computational pathology in collaboration with Dr. Yonatan Winetraub&rsquo;s lab at Stanford University, as well as on underwater computer vision and color restoration under Dr. Derya Akkaynak. Omri is a recipient of competitive scholarships for outstanding MSc. students in AI and Data Science for two consecutive years, and was awarded NeurIPS 2025 financial support in recognition of his research potential.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:506, Zisapel Building
UID:eventx6a5a287eee3cf10976
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20260506T113000
DTEND;TZID=Asia/Jerusalem:20260506T123000
DTSTAMP;TZID=Asia/Jerusalem:20260506T113000
SUMMARY: ceClub  talk by Rishona Daniels (Technion)  CE-Club: Reservoir Computing with Memristive Neural Networks  at 2026-05-06 11:30:00
DESCRIPTION:Reservoir computing is an emerging recurrent neural network approach that requires simple training. It consists of a reservoir layer, which can be built from a time-varying dynamical system, and a readout layer, which is the only part that is trained. Memristors, particularly volatile memristors, are well-suited for constructing the reservoir because their natural decay and nonlinear response allow them to retain recent input information for a short time and act as leaky integrators. In this talk, I will show how volatile memristors can be used to build reservoirs for two tasks: image recognition and time-series prediction. I will explain how the input must be encoded differently for each task to obtain better reservoir states and improve performance. I will also discuss the main design trade-offs involved in these systems, and analyse how important memristor properties such as decay rate, quantization, and device variability affect accuracy, robustness, and hardware cost. By comparing these two tasks within the same reservoir computing framework, I will highlight practical insights into how memristive reservoir systems can be designed and adapted for different applications.\nMSc seminar. Supervisor: Prof Shahar Kvatinsky
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zisapel building 506 &amp; Zoom
UID:eventx6a5a287eee3dc10977
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20260510T150000
DTEND;TZID=Asia/Jerusalem:20260510T160000
DTSTAMP;TZID=Asia/Jerusalem:20260510T150000
SUMMARY: MSC  talk by Amit Vaisman  Fast and Flexible Zero-Shot Diffusion-Based Image Compression  at 2026-05-10 15:00:00
DESCRIPTION:While zero-shot diffusion-based compression methods have seen significant progress in recent years, they remain notoriously slow and computationally demanding. We present an efficient zero-shot diffusion-based compression method that runs substantially faster than existing methods, while maintaining performance that is on par with the state-of-the-art techniques. Our method builds upon the recently proposed Denoising Diffusion Codebook Models (DDCMs) compression scheme. Specifically, DDCM compresses an image by sequentially choosing the diffusion noise vectors from reproducible random codebooks, guiding the denoiser's output to reconstruct the target image. We modify this framework with Turbo-DDCM, which efficiently combines a large number of noise vectors at each denoising step, thereby significantly reducing the number of required denoising operations. This modification is also coupled with an improved encoding protocol. Furthermore, we introduce two flexible variants of Turbo-DDCM, a priority-aware variant that prioritizes user-specified regions and a distortion-controlled variant that compresses an image based on a target PSNR rather than a target BPP. Comprehensive experiments position Turbo-DDCM as a compelling, practical, and flexible image compression scheme.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 401 &amp; Zoom
UID:eventx6a5a287eee3e810974
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20260512T113000
DTEND;TZID=Asia/Jerusalem:20260512T123000
DTSTAMP;TZID=Asia/Jerusalem:20260512T113000
SUMMARY: pixel-club  talk by Roman Beliy (Weizmann Institute of Science)  Pixel Club: Brain-IT: Image Reconstruction from fMRI via Brain-Interaction Transformer  at 2026-05-12 11:30:00
DESCRIPTION:Reconstructing images seen by people from their fMRI brain recordings provides a non-invasive window into the human brain. Despite recent progress enabled by diffusion models, current methods often lack faithfulness to the actual seen images. We present &ldquo;Brain-IT&rdquo;, a brain-inspired approach that addresses this challenge through a Brain Interaction Transformer (BIT), allowing effective interactions between clusters of functionally-similar brain-voxels. These functional-clusters are shared by all subjects, serving as building blocks for integrating information both within and across brains. All model components are shared by all clusters &amp; subjects, allowing efficient training with a limited amount of data. To guide the image reconstruction, BIT predicts two complementary localized patch-level image features: (i)high-level semantic features which steer the diffusion model toward the correct semantic content of the image; and (ii)low-level structural features which help to initialize the diffusion process with the correct coarse layout of the image. BIT&rsquo;s design enables direct flow of information from brain-voxel clusters to localized image features. Through these principles, our method achieves image reconstructions from fMRI that faithfully reconstruct the seen images, and surpass current SotA approaches both visually and by standard objective metrics. Moreover, with only 1-hour of fMRI data from a new subject, we achieve results comparable to current methods trained on full 40-hour recordings.\nRoman Beliy is a PhD candidate at the Weizmann Institute of Science in the Computer Science Department, under the supervision of Prof. Michal Irani. His work lies at the intersection of machine learning, computer vision, and neuroscience.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:506, Zisapel Building
UID:eventx6a5a287eee3f310981
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20260513T103000
DTEND;TZID=Asia/Jerusalem:20260513T113000
DTSTAMP;TZID=Asia/Jerusalem:20260513T103000
SUMMARY: MSC  talk by Maya Wolff  Bridging Multi-Valued Heuristics and Dimensionality Reduction in Multi-Objective Search  at 2026-05-13 10:30:00
DESCRIPTION:Multi-objective shortest-path (MOSP) algorithms traditionally rely on single-valued heuristics (SVHs), which associate each state with a single admissible cost vector. While SVHs provide safe lower bounds, they fail to capture the trade-off structure of the Pareto frontier and often yield weak search guidance. Multi-valued heuristics (MVHs) address this limitation by mapping states to sets of cost estimates, enabling a richer approximation of possible trade-offs.\nModern MOSP algorithms rely heavily on dimensionality reduction (DR) techniques to efficiently perform dominance checks. However, integrating MVHs with DR introduces subtle correctness challenges. We show that naively combining DR with MVHs destroys the ordering invariants required for DR, leading to unsound and incomplete search. To address this issue, we develop the first theoretical frameworks for safely integrating MVHs with DR.\nFirst, we introduce NAMOA*dr-mvh, a theoretical baseline that restores search correctness by enforcing heuristic consistency. Recognizing the practical limitations of this approach, we then introduce our primary contribution, L-NAMOA*dr-mvh. This algorithm employs a "lazy", optimistic approach to DR, preserving exact correctness with only an admissible MVH by dynamically detecting and repairing local ordering violations. Empirical evaluation across a range of benchmarks demonstrates that L-NAMOA*dr-mvh successfully combines the stronger guidance of MVHs with the powerful pruning capabilities of DR, yielding in certain settings a speedup of over 10x compared to existing state-of-the-art MOSP algorithms.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 601
UID:eventx6a5a287eee40010980
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20260513T113000
DTEND;TZID=Asia/Jerusalem:20260513T123000
DTSTAMP;TZID=Asia/Jerusalem:20260513T113000
SUMMARY: ceClub  talk by Or David (Technion)  CE-Club: The Secret Sauce of Distributed Agreement  at 2026-05-13 11:30:00
DESCRIPTION:Agreement problems in general, and Consensus in particular, share fundamental characteristics that have been the subject of extensive research and comparative literature. This work utilizes knowledge theory (epistemic logic) to formally characterize these problems within fault-prone distributed environments. Specifically, we provide a logical characterization of a large class of All-or-Nothing (AoN) agreement problems, including Consensus, Reliable Broadcast, Coordinated Attack and Atomic Commitment. This characterization is built upon the notion of dynamic common knowledge (DCK), a concept recently introduced by Gonczarowski and Moses, which we adapt here for fault-prone distributed systems. By doing so, we offer a new, model-independent approach to the design, analysis, and optimization of agreement protocols. To demonstrate the power and practical utility of this knowledge-theoretic framework, we apply it to two distinct scenarios: optimizing a Consensus protocol in a synchronous crash-failure model, and streamlining Bracha&rsquo;s reliable broadcast protocol in an asynchronous system with Byzantine failures. At the core of this work is the mathematical formalization establishing that DCK is both a necessary and sufficient condition for solving All-or-Nothing agreement. In essence, DCK provides a way to capture the required "state of agreement" among non-faulty processes without forcing them to reach their decisions at the exact same instant. We prove that any protocol satisfying the AoN condition inherently generates this dynamic knowledge. Conversely, we establish a sufficiency theorem: if a system guarantees that a set of facts satisfies Validity (the value is allowed), Exclusivity (no conflicting values are decided), and Epistemic Sufficiency (the fact ensures its own DCK), these facts are entirely sufficient to construct a working agreement protocol. Ultimately, this logical characterization transcends specific network models or failure assumptions, offering a unifying lens through which to view distributed agreement. By demonstrating that attaining DCK is the fundamental epistemic engine driving these protocols, our framework not only demystifies the structure of classical solutions but provides a rigorous, systematic blueprint for designing leaner and more efficient distributed algorithms in the future.\nMSc seminar. Supervisor: Prof Yoram Moses
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zisapel 506 &amp; Zoom
UID:eventx6a5a287eee40c10990
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20260513T123000
DTEND;TZID=Asia/Jerusalem:20260513T143000
DTSTAMP;TZID=Asia/Jerusalem:20260513T123000
SUMMARY: CSpecial Event  DELL Spotlight Day - 13.5.26  at 2026-05-13 12:30:00
DESCRIPTION:We officially open the semester with the first Spotlight Day - and this time DELL is coming to the faculty!\nThis is a great opportunity to get to know the company up close, hear about jobs and career opportunities for students.&nbsp;DELL's recruitment teams will come to meet you, answer questions and tell you about career options at the company.\nWednesday | 13.5 | Tzad starting at 12:30Piano Auditorium, Taub Building\nWe are waiting for you!
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Piano Auditorium, Taub Building
UID:eventx6a5a287eee41a10987
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20260513T170000
DTEND;TZID=Asia/Jerusalem:20260513T180000
DTSTAMP;TZID=Asia/Jerusalem:20260513T170000
SUMMARY: MSC  talk by Daniel Bransky  Named Swapping  at 2026-05-13 17:00:00
DESCRIPTION:A "named page" is a memory page whose content originates from and is backed by a file. Because named pages are regularly read from and written to persistent storage, filesystems strive to preserve file content contiguity, thereby enabling sequential I/O, which can be much faster than random I/O. No analogous effort to preserve contiguity exists for "anonymous pages," which hold unnamed data such as stack or heap bytes. Consequently, swapping a region of anonymous pages in or out can be much slower than reading or writing a region of named pages.\nWe observe (1) that the main advantage of the existing swap mechanism is high swap area utilization, since any anonymous page can be placed at any offset within the swap file, so there is no fragmentation; but (2) that secondary storage is commonly underutilized, so the cost of random I/O may be unwarranted. We therefore propose "named swapping," which associates each anonymous region with its own (swap) file and thus benefits from the underlying filesystem's efforts to maintain contiguity, improving swap performance by up to an order of magnitude. A key challenge we address is anonymous pages shared across multiple regions due to fork-based copy-on-write.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 601
UID:eventx6a5a287eee42510964
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20260513T173000
DTEND;TZID=Asia/Jerusalem:20260513T183000
DTSTAMP;TZID=Asia/Jerusalem:20260513T173000
SUMMARY: CSpecial Event  Come be Part of the Technion's Capture The Flag (CTF) Group!! 13.5.26  at 2026-05-13 17:30:00
DESCRIPTION:Come be part of the Technion's CTF (Capture The Flag) groupNo previous experience needed &ndash; just curiosity and a desire to learn Everyone is welcome, we are waiting for you\nWednesday | 13.5 at 17:30Taub 1\nRegistration link:&nbsp;https://forms.gle/9Kv8VEgLsKPkMjbC8&nbsp;\nCommunity link: https://chat.whatsapp.com/H6NLynYHOnVGo5yfrgu2AS&nbsp;
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 1
UID:eventx6a5a287eee43110982
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20260514T110000
DTEND;TZID=Asia/Jerusalem:20260514T120000
DTSTAMP;TZID=Asia/Jerusalem:20260514T110000
SUMMARY: pixel-club  talk by Roee M. Francos (Harvard University)  Pixel Club: Resilient Decision-Making for Multi-Robot Systems in the Presence of Adversaries  at 2026-05-14 11:00:00
DESCRIPTION:This talk presents a unified perspective on resilient decision-making for multi-robot systems in environments where adversaries may influence routing, search, and detection tasks. Recently, significant progress in coordination, control, and navigation&nbsp;of multi-robot systems has been achieved, driven&nbsp;primarily&nbsp;by the rapid commercialization of unmanned aerial systems and drones. Yet, real-world deployment remains challenging. Most research assumes cooperative or optimally performing agents, overlooking adversarial or suboptimal behavior due to uncertainty, faults, or environmental disturbances. Resilience to malicious or malfunctioning agents remains a key limitation, as such agents can degrade efficiency and destabilize routing policies, where stability is defined as bounded cost over time. Existing stability guarantees for cooperative fleets collapse when agents deviate from plans, underscoring the need for adversarially aware planning and routing theory in safety-critical applications.\nI will present new theoretical and experimental results on resilient multi-agent policies for coordination, control, and learning under uncertainty and adversarial influence, focusing on routing and traffic management for fleets of aerial and ground agents for which I develop adaptive algorithms that provide provable stability, resilience, and safety guarantees. Finally, I will show how resilient cooperative search strategies address challenges posed by intelligent, coordinated external adversaries, providing provable guarantees for coverage and detection in settings such as search and rescue and pursuit-evasion. I conclude by outlining future research directions at the intersection of safety, learning, and large-scale autonomy.\n\nBio:&nbsp;Roee M. Francos is currently a Computer Science Postdoctoral Fellow at the Robotics, Embedded Autonomy, and Communication Theory (REACT) Lab at Harvard University&nbsp;working with Prof. Stephanie Gil, focusing on development of multi-agent resilient decision-making and coordination algorithms. In 2023, he completed his PhD in Computer Science under supervision of Prof. Freddy Bruckstein, at the Technion-Israel Institute of Technology. He received the B.Sc. in Electrical and Computer Engineering from Ben-Gurion University. His research interests are in multi-agent teamwork, autonomous robotics, intelligent transportation systems, bio-inspired robotics and computer vision, focusing on collaborative algorithms for motion planning of autonomous vehicles&nbsp;and&nbsp;multi-robot learning. Roee is a recipient of the 2023 Robotics Science and Systems (RSS) Pioneers Award&nbsp;and the 2025 IEEE Multi-Robot &amp; Multi-Agent Systems (MRS) Young Pioneer Award.\n
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub Piano Auditurion
UID:eventx6a5a287eee43c10984
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20260518T160000
DTEND;TZID=Asia/Jerusalem:20260518T180000
DTSTAMP;TZID=Asia/Jerusalem:20260518T160000
SUMMARY: CSpecial Event  Graduate Studies Open Day - Winter Semester 2026/27  at 2026-05-18 16:00:00
DESCRIPTION:An Information Session about the Master&rsquo;s program, intended for outstanding Bachelor&rsquo;s graduates.\nThe session will take place on Monday, May 18, 2026 at 4:00 PM, in Auditorium 012, Floor 0, Taub Building.\nRegistration ended on May 16, 2026. For details, please contact Sharon Emunah: sharonem@cs.technion.ac.il&nbsp;
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Auditorium 012, Floor 0
UID:eventx6a5a287eee44c10979
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20260519T103000
DTEND;TZID=Asia/Jerusalem:20260519T113000
DTSTAMP;TZID=Asia/Jerusalem:20260519T103000
SUMMARY: PHD  talk by Tomer Adar  Testing and Approximation Algorithms for Very Large Data Sets  at 2026-05-19 10:30:00
DESCRIPTION:Distribution testing has long been studied in the context of theoretical computer science and statistics. The classical model of distribution testing can be seen as too weak in practice, with even the simplest tasks requiring polynomial (though sublinear) sample complexity. We explore non-classical models tailored to cope with this model's infeasibility for distributions defined over very large data sets. We consider two kinds of models: models that are stronger than the classical one, in which testing can be done more efficiently, and a weaker model, which is better suited for extremely high-dimensional data sets.\nThe stronger model is the conditional sampling model, in which the algorithm can sample the input distribution when conditioned on subsets of the domain. We consider this model as well as a few of its restricted variants, such as the subcube conditional model. We tighten the bounds for a few core algorithmic tasks in these conditional models.\nFinally, we explore the behavior of algorithms in the Huge Object model, which combines the classical distribution testing model with the string testing model to more realistically handle distributions over extremely high-dimensional data sets.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 601
UID:eventx6a5a287eee45710983
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20260519T113000
DTEND;TZID=Asia/Jerusalem:20260519T123000
DTSTAMP;TZID=Asia/Jerusalem:20260519T113000
SUMMARY: pixel-club  talk by Nir Mualem (Ben-Gurion University of the Negev)  Pixel Club: Gaussian Splashing: Direct Volumetric Rendering Underwater  at 2026-05-19 11:30:00
DESCRIPTION:&nbsp;\nUnderwater images are strongly affected by light attenuation and backscatter, making 3D reconstruction, depth estimation, and novel-view rendering much harder than in air. In this talk, I will present Gaussian Splashing, a method that adapts 3D Gaussian Splatting to underwater scenes by integrating an underwater image-formation model directly into the rendering and optimization pipeline. The method reconstructs underwater scenes in minutes and renders novel views in real time, while improving depth, color reconstruction, and visual detail compared with existing approaches. I will discuss the motivation, the main technical ideas, experimental results, and possible applications in underwater robotics, marine science, and visualization.\nNir Mualem is a researcher at Ben-Gurion University of the Negev, working on computer vision, 3D reconstruction, and neural rendering, with a particular interest in imaging and reconstruction under challenging underwater conditions. He is currently also working as a Solution Architect at NVIDIA, focusing on optimization methods.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom
UID:eventx6a5a287eee46210991
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20260520T103000
DTEND;TZID=Asia/Jerusalem:20260520T113000
DTSTAMP;TZID=Asia/Jerusalem:20260520T103000
SUMMARY: MSC  talk by Hadar Kaminsky  Generalizing Multi-Objective Search via Objective-Aggregation Functions  at 2026-05-20 10:30:00
DESCRIPTION:Multi-objective search (MOS) has become essential in robotics, as real-world robotic systems need to simultaneously balance multiple, often conflicting objectives. Recent works explore complex interactions between objectives, leading to problem formulations that do not allow the usage of out-of-the-box state-of-the-art MOS algorithms. In this paper, we suggest a generalized problem formulation that optimizes solution objectives via aggregation functions of hidden (search) objectives. We show that our formulation supports the application of standard MOS algorithms, necessitating only to properly extend several core operations to reflect the specific aggregation functions employed.\nWe demonstrate our approach in several diverse robotics planning problems, spanning motion-planning for navigation, manipulation and planning for medical systems under obstacle uncertainty as well as inspection planning, and route planning with different road types. We solve the problems using state-of-the-art MOS algorithms after properly extending their core operations, and provide empirical evidence that they outperform by orders of magnitude the vanilla versions of the algorithms applied to the same problems but without objective aggregation.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub, 1st Floor, CRL lab
UID:eventx6a5a287eee48b10986
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20260520T113000
DTEND;TZID=Asia/Jerusalem:20260520T123000
DTSTAMP;TZID=Asia/Jerusalem:20260520T113000
SUMMARY: ceClub  talk by Dr. Guy Amir (Cornell University)  CE-Club: A Semantic Approach to Verifying Programmable Networks  at 2026-05-20 11:30:00
DESCRIPTION:\nAs networks become more programmable, they are increasingly built around flexible software components. While this programmability enables new functionality and faster innovation, it also makes network behavior harder to reason about. In this talk, I will present a research agenda that brings ideas from formal methods to programmable networks. In particular, I will present techniques that leverage programmable-network semantics for concurrency safety, traffic monitoring, and failure recovery. More broadly, this work illustrates how semantic foundations can help bring stronger correctness guarantees to modern networked systems.\nBio: Guy Amir is a Postdoctoral Researcher at Cornell University, conducting research at the intersection of formal methods, networking, and systems. He earned his Ph.D. in 2024 from the Hebrew University of Jerusalem, where he studied AI safety, focusing on formally verifying&nbsp;reactive AI systems and interpreting neural&nbsp;networks. He holds an M.Sc. in Computer Science and a B.Sc. in Computational Biology and Computer Science, both from the Hebrew University. He has received Rothschild, Fulbright, AI-Net, and Charles Clore fellowships, as well as an ICML Spotlight and KLA Award.\n\n\n
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zisapel building 506 &amp; Zoom
UID:eventx6a5a287eee49d10993
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20260520T123000
DTEND;TZID=Asia/Jerusalem:20260520T143000
DTSTAMP;TZID=Asia/Jerusalem:20260520T123000
SUMMARY: CSpecial Event  NVIDIA Spotlight Day at the Faculty of Computer Science at the Technion 20.5.26  at 2026-05-20 12:30:00
DESCRIPTION:NVIDIA is coming to meet the students of the Faculty of Computer Science at the Technion!Meet on Campus | Wednesday | May 20, 2026 | 12:30 - 14:30\nWhat's on the program:12:30- Mingling and refreshments with the recruitment team and engineers - Faculty Lobby13:00- Lecture System &amp; Networking Product Engineering - Auditorium Taub 213:45- Panel with engineers Q&amp;A - Auditorium Taub 2\nAlumni registration link: http://nvidia.eightfold.ai/events/candidate?plannedEventId=XR3pKXYX2\nStudent registration link: http://nvidia.eightfold.ai/events/candidate?plannedEventId=d09OKNbNd
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub Lobby and Auditorium Taub 2
UID:eventx6a5a287eee4ab10992
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20260520T130000
DTEND;TZID=Asia/Jerusalem:20260520T140000
DTSTAMP;TZID=Asia/Jerusalem:20260520T130000
SUMMARY: Theory Semina  talk by Yaniv Sadeh (Tel Aviv University)  Theory Seminar: Dynamic Edge Coloring of Forests  at 2026-05-20 13:00:00
DESCRIPTION:In the dynamic edge coloring problem, one has to maintain a graph of maximum degree $\Delta$ with at most $\Delta+c$ colors, given updates to the edges of the graph. An important objective is to minimize the recourse, which is the number of edges being recolored.\nWe study this problem on forests, which is a natural yet nontrivial restriction of the problem. We consider the problem in both incremental (edges are only inserted) and fully dynamic (edges may be deleted) models. In the deterministic setting, we show that the natural greedy algorithm achieves $O(\frac{1}{c + \sqrt{\Delta}})$ amortized recourse in the incremental model, and this is tight up to tie-breaking. In contrast, in a fully dynamic forest greedy can be forced to have $\Omega(\log_\Delta n)$ amortized recourse. To partially alleviate this limitation of greedy, we show an optimal non-greedy algorithm with $O(1)$ amortized recourse for \emph{rooted} fully dynamic forests and $c = \Delta - 2$. In the randomized setting, we give a natural distribution-maintaining algorithm that achieves $\Theta(\frac{1}{\Delta})$ expected amortized recourse in the incremental model and $\Theta(\min \{ \frac{\Delta}{c}, \log_{\Delta} n \})$ expected recourse in the dynamic model. These randomized results are optimal for $c=0$.Based on joint work with David Naori and Haim Kaplan.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 401
UID:eventx6a5a287eee4b610994
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20260526T113000
DTEND;TZID=Asia/Jerusalem:20260526T123000
DTSTAMP;TZID=Asia/Jerusalem:20260526T113000
SUMMARY: pixel-club  talk by Noam Issachar (The Hebrew University of Jerusalem)  Pixel Club: DyPE: Dynamic Position Extrapolation for Ultra High Resolution Diffusion  at 2026-05-26 11:30:00
DESCRIPTION:Diffusion Transformer models generate images with remarkable fidelity, yet training them at ultra-high resolutions is often cost-prohibitive due to the quadratic scaling of self-attention. In this talk, I will present Dynamic Position Extrapolation (DyPE), a training-free method that enables pre-trained diffusion transformers to synthesize images at resolutions far beyond their training data with no additional sampling cost.The core of our approach leverages the spectral progression of the diffusion process, where low-frequency structures converge early and high-frequency details emerge in later stages. We introduce a mechanism to dynamically adjust positional encodings at each step, matching the frequency spectrum to the current stage of the generative process. I will demonstrate how DyPE enables models like FLUX to generate images at extreme scales, up to 16 million pixels, while consistently achieving state-of-the-art fidelity on high-resolution benchmarks. https://noamissachar.github.io/DyPE/&nbsp;&nbsp;Noam is a Phd student in the Hebrew University of Jerusalem under the supervision of Prof. Dani Lischinski and Prof. Raanan Fattal. His reserach interest is visual generative models and their applications.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:506, Zisapel Building
UID:eventx6a5a287eee4c210996
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20260527T113000
DTEND;TZID=Asia/Jerusalem:20260527T123000
DTSTAMP;TZID=Asia/Jerusalem:20260527T113000
SUMMARY: MSC  talk by Shay Ben Azra  Approximating Fractional and Generalized Hypertree Width via Semidefinite Programming  at 2026-05-27 11:30:00
DESCRIPTION:We consider the fractional and generalized hypertree width measures, which capture the ability to decompose a hypergraph into tree-like structures. We present improved polynomial-time and parameterized approximation algorithms to these measures: polynomial-time and parameterized approximations to the fractional measure; and a polynomial-time approximation to the generalized measure. All our results improve upon the corresponding previous best known results of [Korchemna-Lokshtanov-Saurabh-Surianarayanan-Xue FOCS&lsquo;2024]. Our approach boils down to a novel use of semi-definite programming for approximating a coverage type vertex separator problem in hypergraphs, that lies at the core of the divide-and-conquer approach to approximating the fractional hypertree width measure.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 601
UID:eventx6a5a287eee4cd10997
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20260527T123000
DTEND;TZID=Asia/Jerusalem:20260527T143000
DTSTAMP;TZID=Asia/Jerusalem:20260527T123000
SUMMARY: CSpecial Event  Amazon Spotlight Day 27.5.26  at 2026-05-27 12:30:00
DESCRIPTION:Want to get an inside look at the world of AI development at Amazon?We invite you to a Spotlight Day with Amazon engineers and recruiters. An opportunity to get to know the people behind the technologies, hear about the real challenges from the industry, understand where the world of development is going, and discover career and student opportunities at the company.\nWednesday, 27.5 between 12:30-14:30 at Taub (Lobby + Piano Auditorium)\nWhat awaits you?12:30-13:00 - Networking with Amazon's engineers and recruitment teams - Taub Lobby13:00-13:45 - Lecture: Agentic Development at Amazon Dr. Oren Kalinsky Senior Applied Scientist at Amazon, will talk about the next generation of software development using AI Agents, working with autonomous agents, and how leading engineers are already developing in a completely different way, more about the lecture here - Piano Auditorium13:45-14:30 - Presentation of open jobs, career opportunities and continued networking\nRegister here\nAnd of course - refreshments as usual&nbsp;\nLooking forward to seeing you!
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub Building, Floor 0
UID:eventx6a5a287eee4d810999
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20260527T130000
DTEND;TZID=Asia/Jerusalem:20260527T140000
DTSTAMP;TZID=Asia/Jerusalem:20260527T130000
SUMMARY: Theory Semina  talk by Shay Moran  Theory Seminar: On the Geometry of Ranking from Counter Examples  at 2026-05-27 13:00:00
DESCRIPTION:Consider the following learning problem.&nbsp;Alice picks a linear order over n elements, and Bob is trying to reveal it.&nbsp; They interact in rounds: in each round Bob proposes a linear order, and if his guess is incorrect, Alice returns a counterexample - a pair whose relative order is wrong.\n&nbsp;How many rounds can Bob guarantee in the worst case?&nbsp; Does the answer change if the counterexamples are chosen uniformly at random?&nbsp; What can be said for more restricted families of linear orders?&nbsp;For example, suppose the elements are embedded as points x₁,&hellip;,xₙ in ℝᵈ, and the order is determined by their distances to an unknown point x*.\nWe will discuss these questions, present solutions to some of them, and describe several open problems and partial results.\nBased on ongoing joint work with Noga Alon and Shlomo Moran, and on a COLT 2026 paper with Mark Braverman, Roi Livni, Yishay Mansour, and Kobbi Nissim.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 8
UID:eventx6a5a287eee4e411000
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20260527T160000
DTEND;TZID=Asia/Jerusalem:20260527T170000
DTSTAMP;TZID=Asia/Jerusalem:20260527T160000
SUMMARY: MSC  talk by Omer Gotfrid  Input-Transformative Tuning: Temporal Domain Alignment for Locked Time-Series Architectures  at 2026-05-27 16:00:00
DESCRIPTION:This research introduces a novel per-timestep input-space adaptation framework designed for multivariate time-series models with fixed weights, addressing the need for secure deployments where regulatory or technical constraints prevent model fine-tuning. By back-propagating task loss through a frozen backbone using target-domain labels, the method enables effective adaptation in both source-free and source-assisted environments without requiring access to original training data. Evaluated against a convolutional backbone across clinical and sensor-based benchmarks, the approach yields substantial performance gains, significantly outperforming existing end-to-end and test-time adaptation baselines. Notably, the adapter matches or exceeds the performance of models trained natively on target data and exhibits unique architectural portability, allowing a single module to be deployed across different frozen predictors in a zero-shot capacity.\n&nbsp;
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom
UID:eventx6a5a287eee4ef10985
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20260531T150000
DTEND;TZID=Asia/Jerusalem:20260531T170000
DTSTAMP;TZID=Asia/Jerusalem:20260531T150000
SUMMARY: CSpecial Event  Research Day at the Faculty of Computer Science  at 2026-05-31 15:00:00
DESCRIPTION:The Graduate Research Day in the Faculty of Computer Science will be held on May 31, 2026, at 3:00 PM, in the lobby of the Taub Computer Science Building.\nResearch Day is an opportunity for faculty students to present their research through posters to faculty members, industry executives, and students at the Faculty and the Technion College.\nThe research participating in the Research Day will be on various topics: Cryptology and Cyber, Data Centers and Clouds, Graphics, Intelligent Systems and Scientific Computation, Machine Learning and Information Retrieval, Systems and Applications, Testing and Verification, Theory of Computer Science.\nThe presenting researches
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Entrance Floor, Taub Computer Science Building
UID:eventx6a5a287eee4fa10941
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20260602T113000
DTEND;TZID=Asia/Jerusalem:20260602T123000
DTSTAMP;TZID=Asia/Jerusalem:20260602T113000
SUMMARY: pixel-club  talk by Narek Tumanyan (Weizmann Institute of Science)  Pixel Club: Harnessing Foundation Model and Architectural Priors for Long-Range Point Tracking and Dynamic 3D Reconstruction  at 2026-06-02 11:30:00
DESCRIPTION:Foundation models have rapidly become a central substrate for visual understanding and generation. They encode rich semantic and geometric priors that can be harnessed for a wide range of vision problems. This work studies how such priors can be combined with architectural inductive biases of deep neural networks to obtain optimization-based, lightweight yet powerful methods for two challenging tasks: (i) long-range video point tracking and, (ii) dynamic 3D reconstruction from monocular video.The first part focuses on DINO-Tracker, a dense long-range point tracking method that leverages self-supervised DINO-ViT features as a semantic prior. To adapt these features for precise tracking, DINO-Tracker combines test-time optimization with a CNN-based feature refinement model, producing localized and temporally consistent features that support robust tracking across long occlusions. The second part introduces DRoPS, a method for dynamic 3D reconstruction from monocular video. DRoPS combines the power of pre-trained tracking and depth models with a novel Deep Motion Prior -- a CNN-based parameterization of the motion field, achieving state-of-the-art results in monocular dynamic 3D reconstruction.&nbsp;Narek is a final-year PhD student (direct-track) at the Weizmann Institute of Science, advised by Prof. Tali Dekel. Previously, he was a Research Scientist Intern at Meta Reality Labs, advised by Jonathon Luiten. His research interests are in generative AI, 3D/4D reconstruction and tracking, and interpretability of vision foundation models for unveiling novel applications.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:506, Zisapel Building
UID:eventx6a5a287eee50611001
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20260602T160000
DTEND;TZID=Asia/Jerusalem:20260602T170000
DTSTAMP;TZID=Asia/Jerusalem:20260602T160000
SUMMARY: MSC  talk by Gil Kizner  Data-Efficient Dynamic MRI via k-Space Frame Interpolation  at 2026-06-02 16:00:00
DESCRIPTION:Cardiac MRI is clinically valuable but inherently slow, requiring many sequential measurements per frame to build a complete image. In dynamic MRI, where each time-frame must be acquired separately, this becomes especially limiting. This work presents a pipeline built on top of TEAM-PILOT model that learns to interpolate videos directly in the frequency domain, generating phase-consistent intermediate frames and effectively enlarging the training dataset without any new acquisitions. The approach addresses two challenges simultaneously: accelerating scan time and alleviating data scarcity, which is a general bottleneck in medical imaging deep learning. We demonstrate that combining 2-shot acquisition (proportional to the amount of signals) with 4 times temporal densification matches standard 8-shot reconstruction quality across multiple state-of-the-art architectures, achieving a 4 times reduction in scan time with no significant loss in image quality.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 9 &amp; Zoom
UID:eventx6a5a287eee51210989
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20260603T130000
DTEND;TZID=Asia/Jerusalem:20260603T140000
DTSTAMP;TZID=Asia/Jerusalem:20260603T130000
SUMMARY: Theory Semina  talk by Dani Dorfman (Max Planck Institute)  Theory Seminar: Improved Tree Sparsifiers in Near-Linear Time  at 2026-06-03 13:00:00
DESCRIPTION:Graphs are often too complex to handle directly, so a recurring theme in graph algorithms is to replace them by simpler objects that approximately preserve the quantities we care about. In this talk, I will discuss one particularly clean form of graph sparsification: tree sparsifiers. A tree cut-sparsifier replaces a capacitated graph by a single tree that approximately preserves every cut, while a tree flow-sparsifier gives a tree whose routable demands can be transferred back to the original graph with bounded congestion.I will start with the classical line of work on tree sparsifiers. I will then present our improved construction for tree sparsifiers in near-linear time. For an undirected capacitated graph on n vertices, our algorithm constructs a tree cut-sparsifier of quality O(log^2 n log log n), nearly matching the best known polynomial-time quality. Via the flow-cut gap, this also yields a tree flow-sparsifier with congestion approximation O(log^3 n log log n).A key feature of the construction is that it is highly modular. We combine previous ideas in tree sparsification with recent advances in expander decomposition, using the latter almost as a black box. This viewpoint lets us simplify the structure of earlier constructions, improve their guarantees, and implement a "refinement phase" in near-linear time.\n&nbsp;Based on a joint work with&nbsp;Daniel Agassy and Haim Kaplan.\n&nbsp;
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 401
UID:eventx6a5a287eee51d11003
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20260604T130000
DTEND;TZID=Asia/Jerusalem:20260604T140000
DTSTAMP;TZID=Asia/Jerusalem:20260604T130000
SUMMARY: PHD  talk by Barak Hefer Gahtan  Modeling what we cannot afford to learn  at 2026-06-04 13:00:00
DESCRIPTION:Real-world deployment of deep learning violates the assumptions of i.i.d. training data and aggregate-metric evaluation: observations are dependent in time, space, and across users; decision latency is bounded by milliseconds; and the cost of a wrong prediction is rarely symmetric. My doctoral research argues that the structural priors hardest for a deep network to recover from data are precisely the ones that should be encoded into its architecture. The position can be stated in one line: what I cannot afford to learn, I encode.\nThe eleven papers in the dissertation span communication systems, physiological monitoring, physics-informed learning, clinical decision support, and large language models. Across them I make three claims. First, architectural priors generalize when they match the structure of the data: the dissertation traces a continuum from priors in the problem formulation, to priors in features and representations, to priors in the architecture, to priors that are the governing equation. Second, this claim is Pareto rather than monotone: aligned priors dominate where parameters are scarce and are dominated where they are not, a scope condition anchored at a clean empirical crossover in the language-model chapters. Third, aggregate metrics are the wrong audit: calibrated diagnostics (matched effective sample size, effective residual-stream depth, the shuffle gap, cost-conditional thresholds) repeatedly reorder conclusions drawn from AUC, perplexity, and fixed-length comparisons.\nThe seminar develops the first claim through three case studies, one at each depth of the continuum: AARL, a DRL scheduler for 5G millimeter-wave networks (prior in the problem formulation); Lab to Wrist, a neural-ODE and neural-Kalman framework that embeds cardiovascular physics architecturally for heart-rate and oxygen-consumption prediction on wearables (prior in the architecture); and a differentiable Randers-Finsler eikonal solver applied to cross-scene wildfire propagation (prior is the governing PDE itself)
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 3 &amp; Zoom&nbsp;
UID:eventx6a5a287eee52a10995
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20260607T173000
DTEND;TZID=Asia/Jerusalem:20260607T183000
DTSTAMP;TZID=Asia/Jerusalem:20260607T173000
SUMMARY: CSpecial Event  AI & API Workshop 7.6.26  at 2026-06-07 17:30:00
DESCRIPTION:You are invited to participate in the AI ​​&amp; API Workshop\nDuring the workshop, we will discuss the transition from a student mindset to a hacker mindset, developing soft and technical skills for hackathons, and we will also learn through practical examples about using external APIs.\nThe workshop will take place on 7.6 at 17:30 at Taub 7, led by Yinon Goldshtein - AI Researcher at Manifold Labs\nRegister herePlease note - the workshop is open to everyone, not just hackathon participants (registration required)\nSee you,CS HACK Team
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:AI &amp; API
UID:eventx6a5a287eee53711010
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20260609T103000
DTEND;TZID=Asia/Jerusalem:20260609T113000
DTSTAMP;TZID=Asia/Jerusalem:20260609T103000
SUMMARY: colloq  talk by Prof. Noga Alon (Princeton University and Tel Aviv University)  Erdős Problems and speculations about the power of AI models  at 2026-06-09 10:30:00
DESCRIPTION:Paul Erdős was a Hungarian mathematician with a remarkable talent for formulating challenging open problems across many areas&nbsp;of mathematics. Several of these problems have recently been resolved either with the assistance of AI systems or, in some cases, through work carried out independently by such models.&nbsp;I will discuss a number of examples, with particular emphasis on the recent solution of the Unit Distance Problem by an internal model of OpenAI. This problem has long been regarded as one of the central open questions in discrete geometry and was among Erdős&rsquo;s favorite problems.&nbsp;Motivated by these developments, I will also speculate on the future capabilities of AI models and their expected impact on mathematical research.&nbsp;The technical aspects of the lecture will be kept light, and no substantial mathematical background will be assumed
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 1
UID:eventx6a5a287eee54211006
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20260609T153000
DTEND;TZID=Asia/Jerusalem:20260609T163000
DTSTAMP;TZID=Asia/Jerusalem:20260609T153000
SUMMARY: pixel-club  talk by Hadar Cohen (The Andrew and Erna Viterbi Faculty of Electrical & Computer Engineering)  Pixel Club: Towards Extended Field-of-View Wavefront Shaping  at 2026-06-09 15:30:00
DESCRIPTION:Optical imaging of biological tissues at depth presents a significant challenge due to strong light scattering within the tissue, which distorts the wavefront and limits the ability to focus light deep inside the sample. One of the main strategies to address this issue is wavefront shaping (WFS), a powerful technique for deep tissue imaging that physically corrects scattering aberrations to maximize the signal-to-noise ratio. However, its practical utility is severely limited by a highly localized correction field. Current efforts to extend the field of view (FoV) rely on the optical memory effect to tile local corrections, but this two-dimensional approximation degrades rapidly in thick, heterogeneous tissue. In this thesis, we investigate algorithms that could allow WFS corrections to apply to a larger tissue volume. We present a comprehensive roadmap for extended FoV WFS by formulating it as a sparse Transmission Matrix (TM) interpolation problem. Using a rigorous synthetic dataset of 3D tissue structures, we systematically benchmark computational approaches ranging from physical multi-scattering inversions (diffraction tomography) to physics-informed deep learning. We demonstrate that transitioning from 2D memory effects to any 3D volumetric prior yields significant performance gains. Furthermore, we reveal a critical &rdquo;generalization gap&rdquo; in deep wavefront shaping. While physics-informed networks effectively learn visually plausible 3D tissue structures to stabilize TM interpolation in highly scattering regimes, their inductive bias fundamentally struggles to estimate the pseudo-random, high-frequency speckle patterns required for accurate optical phase reconstruction. By characterizing this bottleneck between structural regularization and phase fidelity, our benchmark establishes the necessity of hybrid neuro-physical architectures for generalized deep tissue imaging.MSc student under the supervision of Prof. Anat Levin.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:506, Zisapel Building &amp; Zoom Link
UID:eventx6a5a287eee54c11007
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20260610T103000
DTEND;TZID=Asia/Jerusalem:20260610T113000
DTSTAMP;TZID=Asia/Jerusalem:20260610T103000
SUMMARY: colloq  talk by Ido Guy (Meta & Ben-Gurion University)  From Social Recommendation to Trustworthy AI  at 2026-06-10 10:30:00
DESCRIPTION:People generate rich signals as they communicate, search, review, and transact. My research asks how such signals can be harnessed to connect users with the right information, products, and decisions. In the enterprise, I study how social signals can reveal informal knowledge networks in large organizations. On the web, my work explores how people express and pursue information needs through text, voice, and visual queries, and how product knowledge can be extracted and structured at marketplace scale. My current focus is on trustworthy AI for personalized information access. To illustrate this agenda, the talk will present recent work addressing the challenge of training machine learning models when ground-truth labels are delayed by weeks or months in a non-stationary, adversarial environment. This setting creates a fundamental tension between learning from fresh but incomplete feedback and relying on older, more reliable labels that may no longer reflect the current environment. I will show how we formalize these tradeoffs, develop methods to address them, and demonstrate measurable gains through both offline experiments and online A/B tests on live user traffic.\nBio: Ido Guy is a Senior Research Science Manager at Meta and an Adjunct Full Professor at Ben-Gurion University's Institute for Applied AI Research. His research focuses on information retrieval, recommender systems, personalization, and applied AI. Before Meta, he was Chief Scientist and Applied Research Director at eBay Israel, where he established and led four applied science teams in AI for e-commerce. Earlier, he was a Principal Researcher at Yahoo Labs and an STSM/Manager at IBM Research, where he founded and managed the Social Technologies group. Ido has authored over 100 publications and holds 20 granted patents. His work has received multiple best paper and honorable mention awards, an IBM Corporate Award, and the AI-2000 distinction as a top-100 scholar in information retrieval and recommender systems. He currently serves as Program Co-Chair of the ACM Web Conference 2026 and has been a member of the ACM RecSys Steering Committee since 2012.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Auditorium 012, Floor 0
UID:eventx6a5a287eee55911009
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20260610T130000
DTEND;TZID=Asia/Jerusalem:20260610T140000
DTSTAMP;TZID=Asia/Jerusalem:20260610T130000
SUMMARY: Theory Semina  talk by Tomer Adar (Technion)  Theory Seminar: Instance-optimal estimation of L2-norm  at 2026-06-10 13:00:00
DESCRIPTION:The L2-norm, or collision norm, is a core entity in the analysis of distributions and probabilistic algorithms. Batu and Canonne (FOCS 2017) presented an extensive analysis of algorithmic aspects of the L2-norm and its connection to uniformity testing. However, when it comes to estimating the L2-norm itself, their algorithm is not always optimal compared to the instancespecific second-moment bounds, O(1/(&epsilon;∥&micro;∥2) + t&micro;/&epsilon;2 ), for t&micro; = ∥&micro;∥ 3 3/∥&micro;∥ 4 2 &minus; 1, as stated by Batu (WoLA 2025, open problem session). In this paper, we present an unbiased L2-estimation algorithm whose sample complexity matches the instance-specific second-moment analysis. Additionally, we show that Ω(1/(&epsilon;∥&micro;∥2)+ t&micro;/&epsilon;2 ) is indeed the per-instance lower bound for estimating the norm of a distribution &micro; by sampling (even for non-unbiased estimators).\nhttps://arxiv.org/pdf/2602.21937&nbsp;
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 401
UID:eventx6a5a287eee56611011
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20260611T080000
DTEND;TZID=Asia/Jerusalem:20260612T130000
DTSTAMP;TZID=Asia/Jerusalem:20260611T080000
SUMMARY: CSpecial Event  Registration for the 2026 CS Hackathon - Doing Good is Now Open!  at 2026-06-11 08:00:00
DESCRIPTION:Registration for the 2026 CS Hackathon &ndash; Doing Good is now open!Thursday-Friday | June 11-12, 2026\nThe Faculty of Computer Science and the Student Council invite you to take part in the biggest and most significant event of the year.\nThis year, the hackathon will focus on the worlds of emergencies and will deal with developing innovative solutions to improve decision-making and save lives. At a time when every second is critical &ndash; your development can make the difference.\nJoin us for two days of intensive development and teamwork together with rescue and emergency organizations in Israel.\n140 students will embark on a programming marathon with the aim of developing innovative solutions that will help in dealing with emergency and rescue challenges.30 hours of teamwork, gaining experience, mentoring from first-class mentors, lots of treats and of course cash prizes for the winners.It's time to join, influence and build solutions with real impact that change reality.\nPreliminary meetings (mandatory for participants):\nPractical workshop - Programming AIDevelopment of AI applications led by Guy Tamir AI Evangelist at Intel26.5 at 17:30 on Taub 9 (details and registration below)\nPre-Hackathon meeting- Presentation of the rules of the game and Q&amp;A- Challenge panel with emergency personnel from the field31.5 at 17:30 on Taub 2 (details and registration below)AI &amp; API workshopLed by Yinon Goldstein, a second-year student at the faculty7.6 at 17:30 on Taub 7 (details and registration below)\nHackathon registration - now open!https://www.cshack-technion.com/registration&nbsp;Registration in groups of 3-5 participants | Registration must be made as a group and not as individuals | The number of places is limited. Participation confirmations will be sent by 25.5.2026\nNot sure how to get started? We've got you covered\nLooking for an idea?The full list of challenges is waiting on the websitehttps://www.cshack-technion.com/challenges&nbsp;\nDon't have experience?That's exactly why we've prepared short and practical workshops in AI that will give you real tools for developmenthttps://www.cshack-technion.com/pre-hackathon&nbsp;\nLooking for a group or partners?Join the dedicated group and meet other participantshttps://chat.whatsapp.com/KDC6cNRxh490Wcqv9dbEXO?mode=gi_t&nbsp;\nQuestions?cshackathon@cs.technion.ac.il&nbsp;\nGood luckCS Hackathon Team - Doing Good
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub Building
UID:eventx6a5a287eee57210988
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20260611T103000
DTEND;TZID=Asia/Jerusalem:20260611T113000
DTSTAMP;TZID=Asia/Jerusalem:20260611T103000
SUMMARY: MSC  talk by Razya Ladelsky  FinWhale: Two-Round DAG based Byzantine Atomic Broadcast with Optimal Resiliency  at 2026-06-11 10:30:00
DESCRIPTION:DAG-based Byzantine Fault Tolerant (BFT) protocols provide high-throughput consensus under partial synchrony, but existing DAG protocols still require at least a three-message delay to commit decisions.&nbsp;In contrast, existing Fast-Path BFT protocols can achieve optimal termination with a two-message delay under favorable conditions, though they do not naturally extend to DAGs.\nIn this seminar we present FinWhale, the first DAG-based BFT protocol with a two-message delay fast path.&nbsp;FinWhale extends the Mysticeti protocol with a novel fast-path commit mechanism that safely coexists with the protocol&rsquo;s original slow-path rules.&nbsp;To preserve safety across different local DAG views, we introduce new commit structures based on fast-path evidence blocks that enable validators to combine fast-path and slow-path reasoning consistently.\nFinWhale operates in the partially synchronous model with $n=3f+2p-1$ validators; this matches the known lower-bound for fast Byzantine consensus.&nbsp;The protocol tolerates up to $f$ Byzantine faults and achieves fast termination whenever at most $p$ validators fail during the fast path ($1 \leq p\leq f$).Our results show that optimal-latency fast paths can be integrated into uncertified DAG consensus protocols.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 601
UID:eventx6a5a287eee58411002
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20260611T120000
DTEND;TZID=Asia/Jerusalem:20260611T130000
DTSTAMP;TZID=Asia/Jerusalem:20260611T120000
SUMMARY: MSC  talk by Yaniv Holder  Evolve: An Agentic Framework for Refining and Specializing Caching Middleware  at 2026-06-11 12:00:00
DESCRIPTION:The diversity of workloads, performance metrics, and potential system optimizations makes designing an optimal caching middleware a daunting task. For example, cache replacement policies can exhibit widely varying hit ratios depending on workload characteristics and cache sizes. Moreover, the choice of cache replacement policy may impact the computational complexity and meta-data size required for cache management, which may also be impacted by the internal cache organization (e.g., fully associative vs. k-way) and supporting data structures. Consequently, state-of-the-art caching libraries are still hand-crafted by experts and carefully tuned to narrow workload classes and design goals. Adapting a cache to a new setting, or improving it as a robust general-purpose solution, still requires a labor-intensive redesign process that does not scale to the diversity of modern workloads and deployment constraints. We present Evolve, an agentic framework for automatically improving caching middleware implementations.\nStarting from an existing design, Evolve searches for superior variants under two regimes: refinement over a fixed workload suite, and workload specialization against a deployment-specific trace distribution. Each cycle proposes a cohort of LLM-implemented variants, validates them through deterministic correctness gates, scores survivors on a multi-dimensional plugin-based fitness vector, retains a Pareto front under a held-out trace suite to control for overfitting.\nThe framework also stores LLM-generated insights in a knowledge ledger that supports cross-iteration learning and guides future exploration. As a case study, we apply Evolve on a Dash-inspired associative cache, run it across diverse traces in both regimes, and characterize the variants it produces along each chosen fitness dimension.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 601 &amp; Zoom
UID:eventx6a5a287eee59011004
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20260615T110000
DTEND;TZID=Asia/Jerusalem:20260615T120000
DTSTAMP;TZID=Asia/Jerusalem:20260615T110000
SUMMARY: PHD  talk by Omer Rappoport  Constrained Horn Clauses and Theory-Modular Reasoning for Bit-Precise Verification  at 2026-06-15 11:00:00
DESCRIPTION:Constrained Horn Clauses (CHCs) have emerged as a powerful framework for formal verification, enabling reasoning about program safety, invariant synthesis, model checking, and related verification tasks. Their combination of logical expressiveness and algorithmic solvability has made CHCs a central foundation for modern verification techniques. However, despite significant advances in CHC solving, reasoning about low-level bit-precise program semantics remains a major challenge for existing solvers.\nTraditionally, a set of CHCs is encoded with respect to a single background theory. In the context of bit-precise verification, program semantics can naturally be encoded as CHCs modulo the theory of fixed-size bit-vectors. Alternatively, bit-precise semantics can be encoded using integer arithmetic by modeling modular and bit-wise behavior through arithmetic constraints. However, neither approach consistently yields an efficient verification procedure: reasoning directly in the bit-vector theory often limits the scalability and generalization capabilities of CHC solvers, whereas arithmetic encodings produce complex constraints that are expensive to process, especially in the presence of bit-wise operations.\nTo address this limitation, I will present a novel theory-modular framework for CHC satisfiability that enables coordinated reasoning across multiple background theories. The framework decomposes a CHC system into complementary fragments interpreted over different background theories and coordinates reasoning between them through sound cross-theory transformations. This modular approach combines efficient arithmetic reasoning with precise handling of bit-level operations, leveraging the strengths of both theories within a unified CHC-solving framework.\nI will describe the theoretical foundations of the approach, including theory transformations, modular CHC encodings, and a satisfiability procedure for coordinated reasoning across theories. I will also present experimental results on bit-manipulating verification benchmarks, demonstrating substantial improvements over monolithic single-theory approaches. Finally, I will discuss how theory-modular reasoning can contribute to future directions in scalable and expressive CHC solving.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 601
UID:eventx6a5a287eee59d10998
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20260615T140000
DTEND;TZID=Asia/Jerusalem:20260615T150000
DTSTAMP;TZID=Asia/Jerusalem:20260615T140000
SUMMARY: MSC  talk by Tomer Bitan  Guidance and Adaptation Methods for Improving Large Language Models on Parallel Code Translation  at 2026-06-15 14:00:00
DESCRIPTION:High-performance computing software must be ported among serial C++, OpenMP, and CUDA as platforms and deployment constraints change. Such translation requires preserving program behavior while satisfying target-specific constraints on parallel execution, synchronization, memory movement, and API use. Although large language models are flexible code generators, plausible parallel-code translations can fail at compilation, execution, or validation.\nWe study guidance and adaptation methods that make existing LLMs more reliable without training a generator from scratch. The central method is latent PRM guidance, which guides a frozen latent-reasoning model before code is emitted. A small process reward model scores alternative continuous hidden-state branches during latent reasoning and selects branches expected to lead to executable, behaviorally correct programs. The evaluation also includes lighter-weight assistance methods such as prompting, fine-tuning, and iterative repair with compiler and execution feedback.\nOn the ParaTrans dataset, LLMs struggle as standalone executable translators, but their reliability improves significantly through task-aware assistance. Latent PRM guidance improves validation over unguided latent reasoning while remaining compatible with iterative repair. The best results are achieved through the cumulative and compounding effects of combining guidance, prompting, and iterative repair to improve executable correctness and validation performance.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 401 &amp; Zoom
UID:eventx6a5a287eee5ab11008
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20260616T113000
DTEND;TZID=Asia/Jerusalem:20260616T123000
DTSTAMP;TZID=Asia/Jerusalem:20260616T113000
SUMMARY: pixel-club  talk by Omer Dahary (Tel-Aviv University)  Pixel Club: Toward Meaningful Diversity in Text-to-Image Models  at 2026-06-16 11:30:00
DESCRIPTION:Modern text-to-image models achieve strong visual fidelity and prompt alignment, but often at the cost of generative diversity. In this talk, I will present two complementary approaches to this problem: a simple inference-time method that achieves rich diversity by intervening in the model&rsquo;s internal representations, and a new task formulation of controlled diversity, where users explore structured image galleries through meaningful semantic variations.\nOmer Dahary is a PhD student in Computer Science at Tel Aviv University, advised by Daniel Cohen-Or. His research focuses on generative models, with an emphasis on improving their ability to align with user control. He received dual B.Sc. degrees in Computer Science and Mathematics from the Technion through the Rothschild Excellence Program, and subsequently earned his M.Sc. at the Technion under the supervision of Alex Bronstein.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:506, Zisapel Building
UID:eventx6a5a287eee5b611014
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20260617T130000
DTEND;TZID=Asia/Jerusalem:20260617T140000
DTSTAMP;TZID=Asia/Jerusalem:20260617T130000
SUMMARY: Theory Semina  talk by Elizaveta Nesterova (Technion)  Theory Seminar: Optimal Reconstruction from Noisy Linear Queries  at 2026-06-17 13:00:00
DESCRIPTION:In a linear reconstruction game, an unknown point x* &isin; ℝᵈ is estimated from noisy linear queries. In each round, we choose a direction v and receive an answer within &delta; of the inner product ⟨v,x*⟩. The goal is to understand the best worst-case error achievable after T queries.\nI will explain how this interaction is connected to the geometry of ℝᵈ. After T queries, the possible locations of x* consistent with the answers form a feasible region, and the optimal worst-case error is governed by the radius of its minimum enclosing ball. This brings in a classical relation between diameter and minimum-enclosing-ball radius, given by Jung&rsquo;s theorem (1901).\nBuilding on this viewpoint, I will present our main results. In the limit of infinitely many queries, the optimal error is exactly &radic;(2d/(d+1))&delta;. In fixed dimension, the excess error above this limiting value decays doubly exponentially fast in the number of queries. When the dimension grows, there is an exponential query threshold: exponentially many queries are necessary and sufficient to make the excess error vanish.\nThe main technical ingredient is a robust version of Jung&rsquo;s theorem, showing that near-extremal feasible regions have their extremal parts clustered near a regular simplex. I will discuss how this structure leads to the doubly exponential rate, and, time permitting, sketch the proof of the robust Jung theorem.The talk is based on joint work with Yuval Filmus and Shay Moran, accepted to COLT 2026.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 401
UID:eventx6a5a287eee5c111017
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20260621T093000
DTEND;TZID=Asia/Jerusalem:20260621T140000
DTSTAMP;TZID=Asia/Jerusalem:20260621T093000
SUMMARY: CSpecial Event  Google for Science: From Theory to Practice  at 2026-06-21 09:30:00
DESCRIPTION:Google, in collaboration with the Faculty of Computer Science and the Faculty of Electrical and Computer Engineering, invite you to a unique workshop that will expose you to Google's latest research, a fascinating dialogue between academia and industry, and practical applications of artificial intelligence in the world of research and teaching.\nSunday, June 21 | 09:30&ndash;14:00 | Auditorium 012, Floor 0, Taub Building, Faculty of Computer Science\nAll details and the full event program appear in the attached agenda.\nTo register, click here
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Auditorium 012, Floor 0, Taub Building, Faculty of Computer Science
UID:eventx6a5a287eee5cd11005
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20260621T150000
DTEND;TZID=Asia/Jerusalem:20260621T160000
DTSTAMP;TZID=Asia/Jerusalem:20260621T150000
SUMMARY: MSC  talk by Bana Sadi  Accuracy as a limited resource  at 2026-06-21 15:00:00
DESCRIPTION:Prediction algorithms are increasingly used to inform decisions about humans, but maximizing accuracy - the standard learning objective - does not necessarily maximize user benefits. Instead, we propose optimizing social welfare, defined as the average gain users receive from correct predictions. Welfare enables to express, and therefore account for, heterogeneity in how much users benefit from accuracy. But since these valuations are private and users can gain from overreporting them, learning must simultaneously elicit truthful values and optimize welfare with respect to them. To this end, we propose a novel learning algorithm that incorporates a truthful auction. We show how to compute allocations and prices efficiently, and bound the number of paying users - which surprisingly is independent of the sample size. We conclude with experiments on real and synthetic data that demonstrate our algorithm and explore the connections between welfare and accuracy.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 301 &amp; Zoom
UID:eventx6a5a287eee5d811013
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20260621T153000
DTEND;TZID=Asia/Jerusalem:20260621T163000
DTSTAMP;TZID=Asia/Jerusalem:20260621T153000
SUMMARY: MSC  talk by Adi Levy  Expected Recovery Time in DNA-based Distributed Storage Systems  at 2026-06-21 15:30:00
DESCRIPTION:We initiate the study of &nbsp;DNA-based distributed storage systems, where information is encoded across multiple DNA data storage containers to achieve robustness against container failures. In this setting, data are distributed over M containers, and the objective is to guarantee that the contents of any failed container can be reliably reconstructed from the surviving ones. Unlike classical distributed storage systems, DNA data storage containers are fundamentally constrained by sequencing technology, since each read operation yields the content of &nbsp;a uniformly random sampled strand from the container. Within this framework, we consider several erasure-correcting codes and analyze the expected recovery time of the data stored in a failed container. Our results are obtained by analyzing generalized versions of the classical Coupon Collector's Problem, which may be of independent interest.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 601
UID:eventx6a5a287eee5e311012
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20260622T160000
DTEND;TZID=Asia/Jerusalem:20260622T170000
DTSTAMP;TZID=Asia/Jerusalem:20260622T160000
SUMMARY: MSC  talk by Rotem Green  Acoustic World Models for Active Scene Reconstruction  at 2026-06-22 16:00:00
DESCRIPTION:Room impulse responses provide an indirect acoustic probe of scene geometry: as an agent moves, the recorded reverberation changes with nearby walls, openings, and free space. However, converting acoustic observations into maps is inherently ambiguous, since different reflector configurations can produce similar responses, especially when observations are sparse or motion is limited.&nbsp;\nWe study active acoustic scene reconstruction, where an agent must choose sensing poses that improve its geometric belief rather than follow a predefined scan. We introduce a cross-modal acoustic world model that encodes histories of RIRs and known poses into a motion-conditioned latent state used for both local occupancy decoding and future acoustic-latent prediction. At test time, candidate trajectories are rolled out in latent space, decoded into imagined occupancy maps, and scored by predicted map-space information gain. We construct a synthetic benchmark of paired acoustic trajectories, poses, and floor-plan geometry. Experiments show improved local acoustic-to-geometry reconstruction over geometric and passive baselines, and closed-loop mapping that matches or improves frontier-based exploration while substantially reducing collisions.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom
UID:eventx6a5a287eee5ed11016
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20260622T173000
DTEND;TZID=Asia/Jerusalem:20260622T183000
DTSTAMP;TZID=Asia/Jerusalem:20260622T173000
SUMMARY: CSpecial Event  talk by Dror Porat (Video Architect at Nvidia)  Prompt Engineering Workshop: The Key to Effective Use of Generative AI  at 2026-06-22 17:30:00
DESCRIPTION:You are invited to the Prompt Engineering Workshop: The Key to Effective Use of Generative AI\nIn an era where artificial intelligence has become an integral part of the worlds of development, research, and technological practice, the ability to work correctly with AI tools is no longer an advantage, but a basic skill for anyone who wants to remain relevant and lead.\nIf you work with AI tools like ChatGPT and Claude and want to understand how to use them more accurately and effectively, this workshop is for you!\nIn an era where large language models (LLMs) like ChatGPT and Claude are becoming standard tools, the ability to "talk" to the machine is becoming a critical skill (The New Programming Language).\nThis workshop will introduce the principles behind creating effective prompts. We will review advanced methodologies such as Chain-of-Thought, Few-Shot Prompting, and the use of Roles, and we will understand how the structure of the sentence affects the model's probability space.\nThe workshop will combine a brief overview of how to build and train large language models, with a central part in which we will perform an interactive live demo together to solve real-time practical problems from the worlds of research, programming, and data science\nWhat you will get in the workshop&bull; A deep understanding of how the LLM works&bull; A toolbox of prompt templates for effective use in LLMs&bull; The art of context: how to give AI memory and identity&bull; Co-building "on wet" AI solutions for a variety of practical professional tasks\nSpeaker: Dror Porat, Video Architect at NvidiaMonday, June 22 at 5:30 PM at Taub 5\nRegister here, limited number of places\nLet's discover how to turn artificial intelligence from a mere useful tool - into a real growth engine.\nWe're waiting for you!
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 5
UID:eventx6a5a287eee5f911019
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20260623T113000
DTEND;TZID=Asia/Jerusalem:20260623T123000
DTSTAMP;TZID=Asia/Jerusalem:20260623T113000
SUMMARY: pixel-club  talk by Yam Kushinsky (Weizmann Institute of Science)  Pixel Club: Integrating Spherical Harmonic and SplitSum Rendering of Specular Materials  at 2026-06-23 11:30:00
DESCRIPTION:Accurate real-time rendering under complex illumination remains a central challenge in computer graphics, particularly for interactive applications such as gaming. While image-based lighting (IBL) frameworks and the widely used SplitSum approximation enable efficient rendering of specular materials, they rely on simplifying assumptions that introduce artifacts, especially at grazing angles. In this work, we present a novel rendering scheme that improves the approximation of specular reflectance by leveraging spherical harmonics (SH) for low-frequency illumination. For the high-frequency components, we retain the efficiency of the SplitSum method. Our approach decomposes Env. Maps into frequency bands and introduces an analytic SH-based solution to the rendering equation in a unique local shading frame (ULSF), allowing BRDF representations to be efficiently stored in scene-independent lookup tables.We implement our method in Unity and OpenGL, demonstrating improved accuracy over prior approaches, including reduced light leakage and more faithful rendering of rough materials, while maintaining real-time performance with only marginal overhead compared to SplitSum\nYam Kushinsky is a PhD student at the Weizmann Institute of Science, advised by Prof. Ronen Basri. His research focuses on inverse rendering, Gaussian splatting, and real-time rendering &mdash; with a particular interest in recovering geometry, materials, and lighting from images.\nPrior to his PhD, Yam worked for five years as an algorithms developer at Common Ground, where he focused on inverse rendering for human avatars. He holds a Bachelor&rsquo;s degree in Physics and Electrical Engineering, and a Master&rsquo;s degree in Applied Mathematics from the Weizmann Institute, where he studied convex optimization under the supervision of Prof. Yaron Lipman.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:506, Zisapel Building
UID:eventx6a5a287eee60711020
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20260624T123000
DTEND;TZID=Asia/Jerusalem:20260624T133000
DTSTAMP;TZID=Asia/Jerusalem:20260624T123000
SUMMARY: CSpecial Event  Special Event for Students with Camtek  at 2026-06-24 12:30:00
DESCRIPTION:We are pleased to invite you to a special event with Camtek, one of the leading companies in the semiconductor industry, developing and manufacturing advanced inspection and metrology systems for the global semiconductor market.\nThis is a great opportunity to learn more about the company's work, hear about real-world technological challenges from the industry, and explore the fields of software, algorithms, and AI, alongside exciting career opportunities.\nWednesday | June 24, 2026 | Starting at 12:30 PMGraduate Lounge (2nd Floor)\nProgram\n12:30 PM Registration and refreshments\n12:45&ndash;2:00 PM&nbsp;Engaging talks on AI research and AI software infrastructure\n2:00 PM Refreshments, giveaways, and networking with the Camtek team (A special prize will be raffled among participants.)\nParticipation is subject to advance registration. Register here
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Graduate Lounge (2nd Floor)
UID:eventx6a5a287eee61311022
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20260624T130000
DTEND;TZID=Asia/Jerusalem:20260624T140000
DTSTAMP;TZID=Asia/Jerusalem:20260624T130000
SUMMARY: Theory Semina  talk by Guy Arbel (Technion)  Theory Seminar: Determinization of Min-Plus Weighted Automata  at 2026-06-24 13:00:00
DESCRIPTION:Min-Plus Weighted Finite Automata (WFAs) are a quantitative extension of Boolean automata whereby each word is assigned an integer, instead of being accepted or rejected.\nApplications of WFAs fall on a wide spectrum including verification, rewriting systems, tropical algebra, speech and image processing and have been key to proving the star-height conjecture.Unlike Boolean automata, WFAs cannot always be determinized. The decidability of whether a WFA admits an equivalent deterministic WFA is a long standing open problem.\nWe prove that this problem is decidable.As part of the proof, we develop a new toolbox for reasoning about the run structure of weighted automata.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 401
UID:eventx6a5a287eee62111023
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20260630T143000
DTEND;TZID=Asia/Jerusalem:20260630T153000
DTSTAMP;TZID=Asia/Jerusalem:20260630T143000
SUMMARY: colloq  talk by Elias Bareinboim (Columbia University)  Towards Causal Artificial Intelligence  at 2026-06-30 14:30:00
DESCRIPTION:While a significant portion of AI scientists and engineers believe we are on the verge of achieving highly general forms of AI, I offer a critical appraisal of this view through a causal lens. In particular, building on foundational developments in the field, I will present my perspective on the relationship between intelligence and causality, and the central role of the latter in building intelligent systems and advancing credible data science.I frame this discussion in terms of five core capabilities that we should expect from an intelligent AI system: performing causal reasoning and articulating explanations; making precise, surgical, and sample-efficient decisions; generalizing across changing conditions and environments; generating and simulating in a causally consistent manner; and learning causal structures and variables.In this talk, I will elaborate on this perspective and share current progress toward building causally intelligent AI systems. A more detailed discussion of this thesis is provided in my forthcoming textbook, a draft of which is available here: https://causalai-book.net/.\nBio: Elias Bareinboim is a professor in the Department of Computer Science at Columbia University and the director of the Causal Artificial Intelligence (CausalAI) Laboratory. His research develops causality as a foundation for artificial intelligence, with contributions to causal and counterfactual reasoning, data fusion, learning, generalizability, and decision-making, along with applications in biomedical and social domains. His honors include AAAI Fellow recognition, IEEE "AI's 10 to Watch," and young-investigator awards from NSF, DARPA, and ONR. Bareinboim serves as editor-in-chief of the Journal of Causal Inference and as an action editor of the Journal of Machine Learning Research.\nTechnion Host:&nbsp;Sarah Keren
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 337
UID:eventx6a5a287eee62c11021
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20260630T173000
DTEND;TZID=Asia/Jerusalem:20260630T183000
DTSTAMP;TZID=Asia/Jerusalem:20260630T173000
SUMMARY: CSpecial Event  Special CTF Session – Your Home is Your Fortress: How to Protect Your Home Network  at 2026-06-30 17:30:00
DESCRIPTION:You are invited to join the CTF session where we will host a fascinating lecture on the journey to discover and protect personal information online:\nHow do you protect your home network using open source solutions?\nHow do you properly use password managers and multi-factor authentication (MFA)?\nHow do you identify "rogue" devices within your home network?\nWhat is the connection between DNS and your privacy?\nSpeaker: Uri Bar, Security Research Group Manager in the Secure Computing Group at Intel\n**The lecture is also intended for non-IT professionals and does not require in-depth knowledge of networks, but will provide useful tools and insights for anyone who wants to better understand how to protect themselves in the digital world.\nRemember: You are not paranoid &ndash; everyone really wants your data\nTuesday, June 30th at 5:30 PM at Taub 9\nThe meeting is open to everyone and requires pre-registration: https://forms.gle/jTbrANfsMqUSoPbG7&nbsp;\nWe look forward to seeing you!
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:
UID:eventx6a5a287eee63a11027
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20260701T130000
DTEND;TZID=Asia/Jerusalem:20260701T140000
DTSTAMP;TZID=Asia/Jerusalem:20260701T130000
SUMMARY: Theory Semina  talk by Arnold Filtser (Bar Ilan University)  Theory Seminar: How to Protect Yourself from Threatening Skeletons: Optimal Padded Decompositions for Minor-Free Graphs  at 2026-07-01 13:00:00
DESCRIPTION:A metric space has padding parameter &nbsp; if, roughly, for any scale &nbsp;, it admits a stochastic decomposition into clusters of diameter at most &nbsp;, such that any ball of radius &nbsp; is contained within a single cluster with probability at least &nbsp;. The padding parameter is an important characteristic of metric spaces with vast algorithmic implications.This talk will begin by surveying different classic constructions of padded decompositions. We will then discuss the main ideas in our recent proof that the shortest path metric of every &nbsp;-minor-free graph has a padding parameter of &nbsp; (which is also tight). This resolves a long-standing open question and exponentially improves the previous bound.\nA joint work with Jonathan Conroy.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 401
UID:eventx6a5a287eee64611028
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20260701T143000
DTEND;TZID=Asia/Jerusalem:20260701T153000
DTSTAMP;TZID=Asia/Jerusalem:20260701T143000
SUMMARY: PHD  talk by Omer Belhasin  Image Classification via Discrete Diffusion Modeling  at 2026-07-01 14:30:00
DESCRIPTION:Selected for an oral presentation at CPRV 2026; Image classification is a well-studied task in computer vision, and yet it remains challenging under high-uncertainty conditions, such as when input images are corrupted or training data are limited. Conventional classification approaches typically train models to directly predict class labels from input images, but this might lead to suboptimal performance in such scenarios. To address this issue, we propose Discrete Diffusion Classification Modeling (DiDiCM), a novel framework that leverages a diffusion-based procedure to model the posterior distribution of class labels conditioned on the input image. DiDiCM supports diffusion-based predictions either on class probabilities or on discrete class labels, providing flexibility in computation and memory trade-offs. We conduct a comprehensive empirical study demonstrating the superior performance of DiDiCM over standard classifiers, showing that a few diffusion iterations achieve higher classification accuracy on the ImageNet dataset compared to baselines, with accuracy gains increasing as the task becomes more challenging.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom
UID:eventx6a5a287eee65111015
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20260701T143000
DTEND;TZID=Asia/Jerusalem:20260701T153000
DTSTAMP;TZID=Asia/Jerusalem:20260701T143000
SUMMARY: pixel-club  talk by Yulin Wang (School of Biomedical Engineering, ShanghaiTech University)  Pixel Club: Generative AI and Foundation Models for Medical Image Synthesis and Analysis  at 2026-07-01 14:30:00
DESCRIPTION:High-quality multimodal neuroimaging provides complementary information essential to both neuroscience and neurology. However, due to multifaceted practical limitations, complete imaging data is often unavailable in scenarios such as emergency treatment or routine screening. Generative models offer a promising paradigm to address these bottlenecks by imputing missing modalities, enhancing image quality,&nbsp; harmonizing acquisition domain discrepancies, and simulating diverse disease-relevant appearances. This presentation will explore algorithmic innovations in generative AI, including a variety of model architectures and training strategies, with a focus on their versatile applications across multimodal image generation, cross-field MRI synthesis, image super-resolution, and disease-specific image simulation. Furthermore, we will demonstrate how these synthesized and enhanced images effectively facilitate critical downstream tasks, such as subcortical segmentation, cross-modal registration, and clinical diagnosis, with the ultimate goal of improving patient outcomes.\nYulin Wang is a Postdoctoral Researcher at the School of Biomedical Engineering, ShanghaiTech University. In September 2026, she will transition to a postdoctoral position at Cornell Tech, the joint academic venture of Cornell University and the Technion. Her research interests lie at the intersection of medical image computing and analysis, generative artificial intelligence, and neuroradiology, with a primary focus on developing advanced deep learning frameworks to solve challenging reconstruction, synthesis, and enhancement problems in clinical neuroimaging. Her recent works have been published in top-tier journals and conferences, including Cell Reports Medicine, Medical Physics, Physics in Medicine and Biology, and MICCAI.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:1061, Meyer Building &amp; Zoom&nbsp;
UID:eventx6a5a287eee65c11026
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20260707T113000
DTEND;TZID=Asia/Jerusalem:20260707T123000
DTSTAMP;TZID=Asia/Jerusalem:20260707T113000
SUMMARY: pixel-club  talk by Udi Gal (The Andrew and Erna Viterbi Faculty of Electrical & Computer Engineering)  Pixel Club: Learning-Based Retrieval of Cloud Vertical Velocity from Multi-View Satellite Imagery  at 2026-07-07 11:30:00
DESCRIPTION:Accurate retrieval of cloud geometric and dynamical properties from satellite observations is essential for understanding atmospheric processes and improving weather and climate models. While existing stereoscopic techniques can estimate cloud-top height and large-scale atmospheric motion, recovering dense, cloud-scale vertical velocity fields at tens-of-meters spatial resolution from passive satellite imagery remains a challenging inverse problem because the required atmospheric dynamics are not directly observable.\nThis thesis presents a deep learning framework for estimating physically meaningful atmospheric quantities directly from passive multi-view satellite observations. A comprehensive simulation pipeline was developed by combining large-eddy simulations of shallow cumulus clouds with physically based volumetric Monte Carlo rendering, atmospheric radiometric correction, sensor noise modeling, and image alignment to generate realistic synthetic satellite imagery together with corresponding ground-truth atmospheric fields. These data were used to train a fully convolutional encoder-decoder network operating on two temporally consecutive three-view observations.\nThe proposed framework was evaluated on three retrieval tasks: cloud-top vertical velocity, cloud-top height, and vertical velocity at multiple fixed altitude layers throughout the atmospheric column. Experimental results demonstrate accurate retrieval of cloud-top geometry together with successful estimation of cloud-top and volumetric atmospheric motion at the native 20~m spatial resolution of the simulations. The framework recovers physically meaningful atmospheric velocity fields from passive multi-view observations and successfully infers vertical motion throughout much of the cloud-containing atmosphere using a single network architecture and observational input.\nThe presented results demonstrate that passive multi-view satellite imagery contains sufficient information to support dense retrieval of both geometric and dynamical cloud properties without explicit reconstruction of the three-dimensional cloud structure or intermediate physical modeling. These findings establish the feasibility of learning-based atmospheric retrieval from physically realistic simulations and provide a foundation for future development of operational data-driven cloud retrieval methods.\nUdi Gal is an M.Sc. student under the supervision of Prof. Yoav Schechner.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:506, Zisapel Building (Access via the bridge on the 8th floor of Mayer Building)
UID:eventx6a5a287eee66911032
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20260713T130000
DTEND;TZID=Asia/Jerusalem:20260713T140000
DTSTAMP;TZID=Asia/Jerusalem:20260713T130000
SUMMARY: MSC  talk by Omri Kosary  Computing Forest Degree Realizations with Minimum Cardinality Domination  at 2026-07-13 13:00:00
DESCRIPTION:In the Degree Realization problem with respect to a family P of graphs the input is a&nbsp;non-increasing sequence d = (d1, . . . , dn) of positive integers, and the goal is to decide whether&nbsp;there exists a simple undirected graph G &isin; P, whose degrees correspond to d, i.e., such that&nbsp;deg(G) = d. In this paper we consider the version of Degree Realization in which the&nbsp;realization is required to be a forest (i.e., P is the family for forests).&nbsp;We consider optimized Degree Realization in which the goal is to obtain a realization&nbsp;that minimizes an objective function f. That is, the goal is to find a realization G that minimizes&nbsp;f(G) among the realizations of the given input sequence. More specifically, we focus on the&nbsp;following functions: the size of an optimal vertex cover and the size of an optimal dominating&nbsp;set. We also consider the total and paired versions of both Min Vertex Cover and Min&nbsp;Dominating Set. We provide characterizations and linear time realization algorithms for all&nbsp;the above-mentioned problems.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 9
UID:eventx6a5a287eee67711024
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20260715T110000
DTEND;TZID=Asia/Jerusalem:20260715T120000
DTSTAMP;TZID=Asia/Jerusalem:20260715T110000
SUMMARY: MSC  talk by Yuval David  Predicting Cardiovascular Drift from 3D Running Pose via Spatio-Temporal Graph Attention  at 2026-07-15 11:00:00
DESCRIPTION:Cardiovascular drift, the progressive rise in heart rate during sustained exercise at stable intensity, is a sensitive indicator of physiological strain that traditionally requires continuous physiological instrumentation. We show that this latent quantity can be estimated from video alone, with no physiological sensor required at inference time. We propose AsymSGAT, a compact spatio-temporal graph attention network that predicts cumulative cardiac drift from 3D running pose under a session-calibrated protocol: one baseline lap of video establishes a personalized pose reference, after which drift on subsequent laps is predicted directly from changes in running dynamics. The model operates on 3D pose sequences and uses graph-based spatial reasoning over body joints together with temporal attention to capture motion patterns associated with fatigue and cardiovascular load. This seminar will present the motivation, data collection protocol, pose-based modeling approach, experimental setup, and results, demonstrating how video-driven multimodal learning can support non-invasive estimation of physiological dynamics during running.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 601 &amp; Zoom
UID:eventx6a5a287eee68211031
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20260715T113000
DTEND;TZID=Asia/Jerusalem:20260715T123000
DTSTAMP;TZID=Asia/Jerusalem:20260715T113000
SUMMARY: ceClub  talk by Ofir Shapira (Technion)  CE-Club: Enlighten the Black-box: Fuzzing Control Flow Speculation with μarch Simulation  at 2026-07-15 11:30:00
DESCRIPTION:Existing tools for automatic detection of speculative leakage in commercial CPUs cannot be used for fuzzing indirect control flow instructions such as indirect branches, function calls, and returns. This is a fundamental limitation: fuzzers require exhaustive exploration of all possible control paths, potentially encountered during (mis)speculation. This becomes infeasible when any address can be a branch target.\nWe present Flowvizor, the first fuzzer that enables systematic investigation of Spectre-V2-type leakages in commercial CPUs. Our main insight is that the traditional black-box approach of existing tools can be augmented with the publicly available &micro;arch knowledge of specific CPU components, such as branch predictors. By simulating these components when generating inputs to the fuzzer, Flowvizor achieves an exponential reduction of the fuzzing space down to hardware-reachable execution paths. Flowvizor's coverage improves with the accuracy of the &micro;arch simulation and remains strictly larger than that of Revizor, the state-of-the-art black-box CPU fuzzer.\nFlowvizor automatically unveils a new speculation trigger and a leak in Intel's CPU indirect branch predictor, it independently rediscovers several known leaks, offers new insights into the branch predictor &mu;arch, and corroborates recently published reverse-engineered details about its internals. It opens new opportunities for systematic refinement of the reverse-engineered microarchitectural structures, providing code examples that cannot be explained by the public knowledge of CPU internals.\n&nbsp;\nMSc seminar. Supervisor:&nbsp; Prof. Mark Silberstein
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zisapel 506
UID:eventx6a5a287eee68d11036
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20260715T113000
DTEND;TZID=Asia/Jerusalem:20260715T123000
DTSTAMP;TZID=Asia/Jerusalem:20260715T113000
SUMMARY: MSC  talk by Ofir Shapira  Enlighten the Black-box: Fuzzing Control Flow Speculation with μarch Simulation  at 2026-07-15 11:30:00
DESCRIPTION:Existing tools for automatic detection of speculative leakage in commercial CPUs cannot be used for fuzzing indirect control flow instructions such as indirect branches, function calls, and returns. This is a fundamental limitation: fuzzers require exhaustive exploration of all possible control paths, potentially encountered during (mis)speculation. This becomes infeasible when any address can be a branch target.\nWe present Flowvizor, the first fuzzer that enables systematic investigation of Spectre-V2-type leakages in commercial CPUs. Our main insight is that the traditional black-box approach of existing tools can be augmented with the publicly available &micro;arch knowledge of specific CPU components, such as branch predictors. By simulating these components when generating inputs to the fuzzer, Flowvizor achieves an exponential reduction of the fuzzing space down to hardware-reachable execution paths. Flowvizor's coverage improves with the accuracy of the &micro;arch simulation and remains strictly larger than that of Revizor, the state-of-the-art black-box CPU fuzzer.\nFlowvizor automatically unveils a new speculation trigger and a leak in Intel's CPU indirect branch predictor, it independently rediscovers several known leaks, offers new insights into the branch predictor &mu;arch, and corroborates recently published reverse-engineered details about its internals. It opens new opportunities for systematic refinement of the reverse-engineered microarchitectural structures, providing code examples that cannot be explained by the public knowledge of CPU internals.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zisapel 506
UID:eventx6a5a287eee69911037
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20260715T130000
DTEND;TZID=Asia/Jerusalem:20260715T140000
DTSTAMP;TZID=Asia/Jerusalem:20260715T130000
SUMMARY: MSC  talk by Gilad Freidkin  Universal Perturbation Distillation from Individual Adversarial Attacks  at 2026-07-15 13:00:00
DESCRIPTION:Deep Neural Networks remain highly susceptible to perturbation-based attacks, which seek small input modifications that induce model failure. These attacks manifest as either individual or universal adversarial perturbations (IAPs, UAPs), where the former are designed for specific inputs, whereas the latter are input-agnostic. While the simpler setting of IAPs has seen rapid methodological progress, UAP advancements remain comparatively limited, as adapting methods to the universal setting is often nontrivial. In this work, we propose Universal Perturbation Distillation (UPD), a domain-decoupled formulation for learning universal adversarial perturbations from off-the-shelf IAP methods. By treating individual adversarial examples as representation-level supervision, UPD leverages IAP techniques for the universal setting. We instantiate UPD on both large language model jailbreak as well as on image classification settings, achieving and often surpassing state-of-the-art performance, with substantial improvements on robust models.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 601 &amp; Zoom
UID:eventx6a5a287eee6a511034
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20260715T130000
DTEND;TZID=Asia/Jerusalem:20260715T140000
DTSTAMP;TZID=Asia/Jerusalem:20260715T130000
SUMMARY: Theory Semina  talk by Nathan Wallheimer (Weizmann Institute)  Theory Seminar: Triangle Detection in H-Free Graphs  at 2026-07-15 13:00:00
DESCRIPTION:We initiate the study of combinatorial algorithms for Triangle Detection in H-free graphs. The goal is to decide if a graph that forbids a fixed pattern H as a subgraph contains a triangle, using only "combinatorial" methods that notably exclude fast matrix multiplication. Our work aims to classify which patterns admit a subcubic speedup, working towards a dichotomy theorem.\nOn the lower bound side, we show that if H is not 3-colorable or contains more than one triangle, the complexity of the problem remains unchanged, and no combinatorial speedup is likely possible. Conversely, we hypothesize that all remaining patterns admit a combinatorial speedup, and we provide a strongly subcubic algorithm for a rich class of "embeddable patterns" which are characterized by admitting a special type of 3-coloring. Our results confirm the dichotomy hypothesis for all patterns of size up to 8.\nFinally, we extend this main result by proving that our dichotomy hypothesis is equivalent to its counterpart in the much broader setting of induced H-free graphs &mdash; a scenario that a priori seems significantly more challenging. The main ingredient in the proof is a reduction from the induced H-free case to the non-induced H'-free case, where H' preserves the structural properties of H that are relevant for the dichotomy, namely 3-colorability and triangle count. A key technical ingredient is a self-reduction to Unique Triangle Detection that preserves the induced H-freeness property, via a new color coding-like reduction.\nJoint work with Amir Abboud and Ron Safier.&nbsp;
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 401
UID:eventx6a5a287eee6b111038
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20260720T123000
DTEND;TZID=Asia/Jerusalem:20260720T133000
DTSTAMP;TZID=Asia/Jerusalem:20260720T123000
SUMMARY: MSC  talk by Oren Hecht  Beyond Program Equivalence: Physics-Aware Search for Molecular Dynamics Optimization  at 2026-07-20 12:30:00
DESCRIPTION:Molecular dynamics simulations are computationally expensive, but optimizing them requires more than preserving exact program behavior. Many useful changes alter numerical trajectories while still preserving the physical properties that matter for a given simulation, such as energy stability, reversibility, or ensemble-level statistics.\nWe present a source-to-source optimization framework that searches for faster molecular dynamics implementations under physics-aware validation. The framework combines equivalence-preserving rewrites with stochastic program mutations that deliberately explore beyond ordinary semantic equivalence. For each candidate, a staged verifier checks both structural requirements and simulation-specific physical behavior, while a population-based search balances runtime, physical deviation, and program simplicity.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 301
UID:eventx6a5a287eee6bc11030
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20260721T113000
DTEND;TZID=Asia/Jerusalem:20260721T123000
DTSTAMP;TZID=Asia/Jerusalem:20260721T113000
SUMMARY: pixel-club  talk by Danielle Sapir (The Andrew and Erna Viterbi Faculty of Electrical & Computer Engineering)  Pixel Club: genZ – Experimental Snapshot 3D Imaging in Fluorescence Microscopy  at 2026-07-21 11:30:00
DESCRIPTION:Three-dimensional (3D) fluorescence microscopy is limited by slow scanning speeds and susceptibility to photobleaching, while faster volumetric imaging modalities frequently impose prohibitive costs or significant resolution tradeoffs. In this seminar, we present an easily adaptable solution for snapshot 3D microscopy via the joint optimization of an axial-encoding Point Spread Function (PSF) and a deep learning-based reconstruction algorithm.\nWe will explore the problem domain and challenges, as well as the building blocks of our solution: designing, collecting, and preprocessing a large-scale 3D microscopy dataset, physics-aware differentiable imaging model, imaging-plane selection module, and model architecture.\nOur method enables single-shot 3D volumetric reconstruction using an engineered depth-encoding PSF, allowing 2D-to-3D dynamic imaging restoration. We will review results in simulated and experimental settings of static and live samples.\nDanielle Sapir is an M.Sc. candidate under the supervision of Prof. Yoav Shechtman.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:506, Zisapel Building &amp; Zoom&nbsp;
UID:eventx6a5a287eee6c711040
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20260722T103000
DTEND;TZID=Asia/Jerusalem:20260722T113000
DTSTAMP;TZID=Asia/Jerusalem:20260722T103000
SUMMARY: MSC  talk by Peer Sagiv  Data-Driven Online Knapsack with Departures and its Applications to Cloud Resource Allocation  at 2026-07-22 10:30:00
DESCRIPTION:In this lecture, we study knapsack problems with departures under the online with a&nbsp;sample model. We begin with the fundamental special case of the Temp Secretary Problem with departures, where we obtain a constant competitive ratio of 1/8, providing the&nbsp;first performance guarantee for general instances of this problem.\nWe then extend our approach to the d-dimensional Online Vector Generalized Assignment Problem with Departures (VGAPWD), achieving a competitive ratio of 1/(16d) for d-dimensional resources. Lastly, we study the Multiple Knapsack Problem With Departures, which is a special case of VGAP.\nFor this special case, we present a more practical modification of our algorithm that achieves the same competitive ratio. Using extensive simulations on workloads derived from real cluster traces, we demonstrate that our algorithms consistently outperform state of the art algorithms and widely used heuristics, achieving typical improvements of 10&ndash;25% in total value compared to state of the art approaches.\nThese results demonstrate that the online with a sample paradigm successfully translates into algorithms that leverage historical data for improved empirical performance. This is aligned with the stronger theoretical guarantees we can prove within this framework.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom
UID:eventx6a5a287eee6d211029
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20260722T123000
DTEND;TZID=Asia/Jerusalem:20260722T143000
DTSTAMP;TZID=Asia/Jerusalem:20260722T123000
SUMMARY: CSpecial Event  The Annual Project Fair Of The Taub Faculty Of Computer Science  at 2026-07-22 12:30:00
DESCRIPTION:We are pleased to invite you to&nbsp;the annual project fair&nbsp;of the Taub Faculty of Computer Science along with the&nbsp;Outstanding Project Competition - Wednesday, July 22, starting at 12:30 in the Taub Lobby - Floor 0.\nDuring the fair, you will have the opportunity to explore a wide range of innovative projects developed by our students, including:&bull; A distributed parking management system integrating computer vision&bull; A user-friendly, game-based hand rehabilitation system&bull; A system for analyzing the relationship between walking pace and music tempo as a foundation for clinical research on neurological disorder rehabilitation&bull; A system for reserving accessible parking spaces for people with disabilities to improve campus accessibility&bull; An upgraded management system for abandoned properties in Haifa, developed in collaboration with the Haifa Municipality&bull; AI-powered optimization of logistics chain management in an advanced pharmaceutical manufacturing facility&bull; And many more exciting and inspiring projects\nThis is a great opportunity to experience firsthand the creativity, innovation, and development capabilities of our students, support the teams competing for the Outstanding Project Award, and discover technological solutions with real-world impact.\nList of projects showing in the link\nWe are waiting for you!
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub Lobby, Floor 0
UID:eventx6a5a287eee6e911018
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20260727T163000
DTEND;TZID=Asia/Jerusalem:20260727T173000
DTSTAMP;TZID=Asia/Jerusalem:20260727T163000
SUMMARY: PHD  talk by Yakir Yehuda  Clinical Classification of multivariate time series  at 2026-07-27 16:30:00
DESCRIPTION:Clinical multivariate time series, such as 12-lead electrocardiograms provide essential information for diagnosis and patient monitoring. However, learning reliable models for these signals remains challenging because they exhibit complex temporal dynamics, strong dependencies between channels, limited labeled data, and physiological constraints that are often ignored by standard deep learning methods.\nWe address these challenges by introducing structure-aware learning approaches for clinical time-series classification and generation. Our main motivation is that existing models often treat multichannel physiological signals as generic data, without explicitly modeling their dynamical behavior, physical structure, or inter-lead relationships. To overcome this limitation, we explore several complementary directions: using Koopman-based temporal dynamics for self-supervised representation learning, incorporating ODE-based cardiac simulators into generative models, designing PDE-driven architectures that learn spatiotemporal representations of 12-lead ECGs, and integrating physiological priors into diffusion models for ECG generation.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom
UID:eventx6a5a287eee6f611041
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20260730T140000
DTEND;TZID=Asia/Jerusalem:20260730T150000
DTSTAMP;TZID=Asia/Jerusalem:20260730T140000
SUMMARY: PHD  talk by Zorik Gekhhman  Understanding Hallucinations in LLMs Through Hypothesis-Driven Research  at 2026-07-30 14:00:00
DESCRIPTION:Factual reliability remains a critical bottleneck for LLMs, whose tendency to hallucinate undermines user trust. To mitigate this problem, we need methods to scientifically measure what a model knows and why it makes factual mistakes. However, many fundamental research questions in this space are not trivial to test, demanding the design of dedicated, controlled experiments precisely tailored to isolate the relevant phenomenon. Using three recent papers as case studies, we will explore the full research cycle: from defining the research question and formulating a hypothesis, to designing a controlled study to test it.&nbsp;\nThe papers we will discuss are:[1] Does Fine-Tuning LLMs on New Knowledge Encourage Hallucinations? (EMNLP 2024)[2] Inside-Out: Hidden Factual Knowledge in LLMs (COLM 2025)[3] Thinking to Recall: How Reasoning Unlocks Parametric Knowledge in LLMs (COLM 2026)
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Bloomfield 526
UID:eventx6a5a287eee70011039
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID=Asia/Jerusalem:20260802T123000
DTEND;TZID=Asia/Jerusalem:20260802T133000
DTSTAMP;TZID=Asia/Jerusalem:20260802T123000
SUMMARY: MSC  talk by Ido Ram  Narrower is Better: Task-Aware API Transpiling for Software Agents  at 2026-08-02 12:30:00
DESCRIPTION:Modern software engineering agents solve tasks by interleaving reasoning with API calls that observe and modify their environment. On long-horizon tasks this degrades performance: the context fills with API documentation, authentication flows, and parameter details, forcing the agent to interleave reasoning about the task with discovery of the interface used to solve it, a conflict that compounds into failure.\nWe introduce SCOPE, an approach that decouples task analysis from implementation by a method of API transpiling. A planner agent first reasons about the task in isolation, defines a focused set of environment interactions it requires, and transpiles the raw APIs into a compact, task-specific toolset, verifying each tool before handoff. A solver agent then operates over this narrowedinterface, free of extraneous API documentation and complicated workflows.\nOn the AppWorld Benchmark with Minimax-M2, SCOPE reaches 81% accuracy on test_normal, a 5-point gain over a vanilla ReAct baseline under identical conditions, and a 71% accuracy on test_challenge, a 6-point gain over the baseline, while shortening solver trajectories by 48% and collapsing the exposed API surface from 74 endpoints to 4 task-specific tools on average.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 9
UID:eventx6a5a287eee70b11035
END:VEVENT
END:VCALENDAR