BEGIN:VCALENDAR
VERSION:2.0
METHOD:PUBLISH
X-WR-CALNAME:IL-Technion-CS-events
PRODID:-//tzurl.org//NONSGML Olson 2010k//EN
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":20210104T160000
DTEND;TZID="Asia/Jerusalem":20210104T170000
DTSTAMP;TZID="Asia/Jerusalem":20210104T160000
FREEBUSY;FBTYPE=BUSY:20210104T160000/20210104T170000
SUMMARY;LANGUAGE=en-US:cggc talk by Ergun Akleman (Texas A&M University) a
bout CGGC Seminar: Hyper-Realistic Rendering: Leveraging Artistic & Mathem
atical Approaches for Effective Control of Visual Results at 2021-01-04 1
6:00:00
DESCRIPTION;LANGUAGE=en-US:My primary goal in this fringe direction of res
earch is to develop a simple, intuitive formal framework for the automatic
representation of simplified shapes and materials that can support Hyper-
Realism in a wide variety of rendering applications. I observe that with t
he emphasis on the physical laws in rendering systems, (1) the focus incre
asingly shifts away from how users perceive the virtual environment, (2) r
endering becomes prohibitively difficult to realize desired global illumin
ation effects in real-time, and (3) the true inclusion of human-in-the-loo
p to control visual results also becomes significantly hard. I have identi
fied two broad categories of artistic and mathematical approaches that can
facilitate effective Hyper-Realistic rendering with clear control of visu
al results: (1) Geometry Representation with Anamorphic Bas-Reliefs and (2
) Material Representation with Barycentric Shaders. A significant advantag
e of these two approaches is that they simplify the reconstruction process
es by allowing some of the real-world parameters to be embedded into the r
epresentations. In this talk, I will give a wide variety of examples that
demonstrate the effectiveness of this approach.
Bio: Ergun Akleman is
a Professor in the Departments of Visualization & Computer Science and En
gineering. Akleman has been at Texas A&M University for 25 years. He recei
ved his Ph.D. degree in Electrical and Computer Engineering from the Georg
ia Institute of Technology in 1992. Akleman is teaching, research and crea
tive activities are all transdisciplinary. He had more than 150 publicatio
ns in a wide variety of disciplines from computer graphics, computer-aided
design, and mathematics to art, architecture, and social sciences. His mo
st significant and influential contributions as a researcher have been in
shape modeling and computer-aided sculpting. He is also a professional car
toonist who published more than 500 cartoons. He has a bi-monthly corner c
alled Computing through Time in the Flagship magazine of IEEE Computer Soc
iety, IEEE Computer.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture: https://technion.zoom.us/j/91344952941
UID:123se2401202435540
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20210105T103000
DTEND;TZID="Asia/Jerusalem":20210105T113000
DTSTAMP;TZID="Asia/Jerusalem":20210105T103000
FREEBUSY;FBTYPE=BUSY:20210105T103000/20210105T113000
SUMMARY;LANGUAGE=en-US:CSpecial Talk talk by Hila Peleg (CSE, University o
f California, San Diego) about CS Lecture: Next Generation Programming wit
h Program Synthesis at 2021-01-05 10:30:00
DESCRIPTION;LANGUAGE=en-US:Program synthesis is the problem of generating
a program to satisfy a specification of user intent. Since these specifica
tions are usually partial, this means searching a space of candidate progr
ams for one that exhibits the desired behavior. The lion's share of the wo
rk on program synthesis focuses on new ways to perform the search, but har
dly any of this research effort has found its way into the hands of users.
We wish to use synthesis to augment the programming process, leverag
ing both optimized search algorithms and concepts that are part of the pro
grammer's life such as code review and read-eval-print loops (REPL). This
talk describes three synthesis-based techniques that bring program synthes
is into the development workflow.
A major concern in designing for th
e user is that it can put the interface of the synthesizer at odds with st
ate of the art synthesis techniques. Synthesis is, at best, a computationa
lly hard problem, and any changes made to make the tool more usable can in
terfere with the synthesizer and its internals. We therefore demonstrate t
he process of bringing synthesis theory into practice when tool design als
o requires an algorithm re-design.
Bio:
Hila Peleg is a postdoctora
l researcher in the CSE Department at University of California, San Diego,
advised by Nadia Polikarpova. She received her PhD at the Technion, Israe
l. She also holds a degree in literature. Her research explores the way pr
ogram synthesis can transform the programming experience.
Zoom Lectur
e:
https://technion.zoom.us/j/96189772408?p
wd=MGF4RVJtNjgxWExMUEhMb1hZWUt6Zz09
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture:
https://technion.zoom.us/j/96189772408?pwd=MGF4RVJtNjgxWExMUEhMb1hZWUt6Zz09
UID:123se2401202435550
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20210105T113000
DTEND;TZID="Asia/Jerusalem":20210105T123000
DTSTAMP;TZID="Asia/Jerusalem":20210105T113000
FREEBUSY;FBTYPE=BUSY:20210105T113000/20210105T123000
SUMMARY;LANGUAGE=en-US:pixel-club talk by Itai Lang (Tel-Aviv University)
about Pixel Club: Learned Sampling of 3D Point Clouds at 2021-01-05 11:30:
00
DESCRIPTION;LANGUAGE=en-US:There is a growing number of tasks that work di
rectly on point clouds. As the size of the point cloud grows, so do the co
mputational demands of these tasks. A possible solution is to sample the p
oint cloud first. Classic sampling approaches, such as farthest point samp
ling (FPS), do not consider the downstream task. A recent work showed that
learning a task-specific sampling can improve results significantly. Howe
ver, the proposed technique did not deal with the non-differentiability of
the sampling operation and offered a workaround instead. We introduce a n
ovel differentiable relaxation for point cloud sampling that approximates
sampled points as a mixture of points in the primary input cloud. Our appr
oximation scheme leads to consistently good results on classification and
geometry reconstruction applications. We also show that the proposed sampl
ing method can be used as a front to a point cloud registration network. T
his is a challenging task since sampling must be consistent across two dif
ferent point clouds for a shared downstream task. In all cases, our approa
ch outperforms existing non-learned and learned sampling alternatives.
The talk will cover the following recent papers:
1. Learning to Sample
, CVPR 2019.
2. SampleNet: Differentiable Point Cloud Sampling, CVPR 202
0 (Oral)
Bio:
I'm a PhD candidate at Tel Aviv University, advised b
y Professor Shai Avidan. My current research interests are computer vision
and learning methods for 3D point clouds. I received my BSc degree in Ele
ctrical Engineering and Physics from the Technion (Summa Cum Laude), as an
alumnus of "Psagot", the elite academic program of the IDF. I hold an MSc
degree in Electrical Engineering from Tel Aviv University, where I studie
d lesion segmentation in medical images.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture: https://technion.zoom.us/j/95495412165
UID:123se2401202435560
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20210107T103000
DTEND;TZID="Asia/Jerusalem":20210107T113000
DTSTAMP;TZID="Asia/Jerusalem":20210107T103000
FREEBUSY;FBTYPE=BUSY:20210107T103000/20210107T113000
SUMMARY;LANGUAGE=en-US:CSpecial Talk talk by Eylon Yogev (Tel-Aviv Univers
ity & Boston University) about CS Lecture: Adversarially Robust Streaming
Algorithms at 2021-01-07 10:30:00
DESCRIPTION;LANGUAGE=en-US:A streaming algorithm is given a long sequence
of items and seeks to compute or approximate some function of this sequenc
e using a small amount of memory. A body of work has been developed over t
he last two decades, resulting in optimal streaming algorithms for a wide
range of problems.
While these algorithms are well-studied, the vast
majority of them are defined and analyzed in the static setting, where the
stream is assumed to be fixed in advance, and only then the randomness of
the algorithm is chosen. In many scenarios, this assumption is unrealisti
c, making the algorithms prone to adversarial attacks, and unfit for real-
life applications.
I will investigate the adversarial robustness of s
treaming algorithms. An algorithm is considered robust if its performance
guarantees hold even if the stream is chosen adaptively by an adversary th
at observes the outputs of the algorithm along the stream and can react in
an online manner. I will describe general-purpose methods we have develop
ed to transform standard streaming algorithms to be secure in the adversar
ial model. These are the first streaming algorithms to work in this model.
Bio:
Eylon Yogev is a postdoc at Tel Aviv University hosted by Ome
r Paneth and Nir Bitasnky and co-affiliated at Boston University hosted by
Ran Canetti. Prior to that, he was a research fellow at the Simons Instit
ute and a postdoc at the Technion hosted by Yuval Ishai. He received his P
h.D. and M.Sc. from the Weizmann Institute of Science under the supervisio
n of Moni Naor and completed his B.Sc. in Computer Science at Tel Aviv Uni
versity. Eylon is broadly interested in cryptography, with a focus on its
interplay with the area of algorithm design and analysis.
Zoom Lectur
e: https://technion.zoom.us/j/99408763465?pwd=
dk9vR2FueGpDbk4vNlJMdm5CU0tmQT09
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture: https://technion.zoom.us/j/99408763465?pwd=dk9vR2FueGpDbk4vNlJMdm5CU0tmQT09
UID:123se2401202435570
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20210110T173000
DTEND;TZID="Asia/Jerusalem":20210110T193000
DTSTAMP;TZID="Asia/Jerusalem":20210110T173000
FREEBUSY;FBTYPE=BUSY:20210110T173000/20210110T193000
SUMMARY;LANGUAGE=en-US:CSpecial Event about Campus Hour by ELBIT at 2021-0
1-10 17:30:00
DESCRIPTION;LANGUAGE=en-US:You are invited to a Campus Day by Elbit System
s,
which presents opportunities and technologies, and a lecture by Yona
tan Avraham,
Development Team Leader, on unique solutions of infrastruc
ture-free
communication, as well as an open conversation with Guy Istma
ti, Director of
Academy Relations, about career opportunities and recru
itment processes.
To
view
register in advance.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Event: Registration
UID:123se24012024100360
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20210111T110000
DTEND;TZID="Asia/Jerusalem":20210111T120000
DTSTAMP;TZID="Asia/Jerusalem":20210111T110000
FREEBUSY;FBTYPE=BUSY:20210111T110000/20210111T120000
SUMMARY;LANGUAGE=en-US:cggc talk by Helmut Pottmann (TU WIEN, Applied Geom
etry) about CGGC Seminar: Quad-mesh Based Mappings between Surfaces at 202
1-01-11 11:00:00
DESCRIPTION;LANGUAGE=en-US:We discretize mappings between surfaces as corr
espondences between checkerboard patterns derived from quad meshes. This m
ethod captures the degrees of freedom inherent in smooth maps and provides
a very simple and efficient computational approach to important types of
maps such as conformal or isometric maps. In particular, it enables a natu
ral definition of discrete developable surfaces which is much more flexibl
e in applications than previous concepts of discrete developable surfaces.
We discuss geometric modeling of developable surfaces, including cutting,
gluing and folding, and present a construction of watertight CAD models c
onsisting of developable spline surfaces. Moreover, we outline further app
lications of quad-mesh based maps in architectural geometry and computatio
nal fabrication.
The lecture will be recorded.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture: https://technion.zoom.us/j/91344952941
UID:123se2401202435600
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20210112T160000
DTEND;TZID="Asia/Jerusalem":20210112T170000
DTSTAMP;TZID="Asia/Jerusalem":20210112T160000
FREEBUSY;FBTYPE=BUSY:20210112T160000/20210112T170000
SUMMARY;LANGUAGE=en-US:CSpecial Talk talk by Nadav Dym (Duke University) a
bout CS Lecture: Computational Theory of Graphs, Sets and Rigid Sets at 20
21-01-12 16:00:00
DESCRIPTION;LANGUAGE=en-US:Quotient spaces are a natural mathematical tool
to describe a variety of algorithmic problems where different objects are
to be compared while their natural symmetries are to be ignored. In parti
cular, we will focus on graphs and sets whose symmetries are permutation o
f the vertices, and rigid sets whose symmetries also include rigid motions
. All three data types are prevalent in computer vision/graphics and in ma
ny other applications.
We will discuss two problems involving these
data types: (1) Geometric alignment of graphs/sets/rigid sets, and whether
it can be done in polynomial time. (2) Neural network architectures which
are invariant to the symmetries of graphs/sets/rigid sets, and whether th
ey are universal (can approximate every invariant continuous function). Fo
r both problems we will see that they are tractable for sets and intractab
le for graphs. We will then explain why rigid sets are an intermediate cas
e, where high dimensional rigid sets are equivalent in graphs in terms of
complexity, while for fixed low dimension they are tractable. The focus on
the lecture will be on two of my recent papers which leverage these insig
hts to achieve tractable algorithms for low-dimensional rigid sets, both f
or geometric alignment and for invariant neural networks.
Bio:
Nada
v Dym is an applied mathematician and computer scientist, interested in th
e development and analysis of algorithms, typically targeting 3D problems
in computer vision and related fields. He is working on problems related t
o global optimization, phase retrieval, and theoretical and geometric deep
learning. He is an Assistant Research Professor at the Department of Math
ematics of Duke University, hosted by Prof. Ingrid Daubechies at the Infor
mation Initiative at Duke (iiD). In 2018 he completed his PhD in the Depar
tment of Computer Science and Applied Mathematics at the Weizmann Institut
e of Science under the supervision of Prof. Yaron Lipman. He received his
BSc and MSc in mathematics from the Hebrew University.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture:
91344952941
Meeting ID: 378 331 9350
Passcode: CSLECTURE
UID:123se2401202435590
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20210114T123000
DTEND;TZID="Asia/Jerusalem":20210114T133000
DTSTAMP;TZID="Asia/Jerusalem":20210114T123000
FREEBUSY;FBTYPE=BUSY:20210114T123000/20210114T133000
SUMMARY;LANGUAGE=en-US:phd talk by Dolev Adas about Concurrent Sketches an
d their Applications at 2021-01-14 12:30:00
DESCRIPTION;LANGUAGE=en-US:Sketches maintain compact approximate statistic
s about streams of data, thereby enabling quickly answering queries regard
ing the data stream without having to reprocess it.
In this talk we w
ill present four different papers that studies concurrent sketches and the
ir applications. In particular we looked at these subjects : CRDT sliding
window sketch, Multi-Producers Single-Consumer Queue, Limited Associativit
y Caches and Cache Admission Filter .
In first result we introduce th
e notion of sliding window CRDT sketches where we collect global statistic
s about the stream in a decentralized manner.
In the second result we
study the implementation of wait-free multi-producer single-consumer queu
es. In applications such as sharded data processing systems, data flow pro
gramming and load sharing applications, multiple concurrent data producers
are feeding requests into the same data consumer. This can be naturally r
ealized through concurrent queues, where each consumer pulls its tasks fro
m its dedicated queue.
In the third result we show that limited assoc
iativity caches are a promising direction for software caches. Specificall
y, we demonstrate that limited associativity enables simple yet efficient
realizations of multiple cache management schemes that can be trivially pa
rallelized. We show that the obtained hit ratio is usually similar to full
y associative caches of the same management policy, but the throughput is
improved by up to x5 compared to production-grade caching libraries, espec
ially in multi-threaded executions.
In the last result we presented T
inyCache - An Effective Cache Admission Filter is a compact table based m
anagement policy for datastore caches that is easily parallelizable.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture:
https://technion.zoom.us/j/94996166729?pwd=L0VOWWJJKytzSkVTT2w1N2FzUzdjUT09
UID:123se2401202435580
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20210114T170000
DTEND;TZID="Asia/Jerusalem":20210114T180000
DTSTAMP;TZID="Asia/Jerusalem":20210114T170000
FREEBUSY;FBTYPE=BUSY:20210114T170000/20210114T180000
SUMMARY;LANGUAGE=en-US:CSpecial Talk talk by Or Litany (NVIDIA, Toronto AI
lab) about CS Lecture: Learning on Pointclouds for 3D Scene Understanding
at 2021-01-14 17:00:00
DESCRIPTION;LANGUAGE=en-US:In this talk i'll be covering several works in
the topic of 3D deep learning on pointclouds for scene understanding tasks
. First, I'll describe VoteNet (ICCV 2019, best paper nomination): a metho
d for object detection from 3D pointclouds input, inspired by the classica
l generalized Hough voting technique. I'll then explain how we integrated
image information into the voting scheme to further boost 3D detection (Im
VoteNet, CVPR 2020). In the second part of my talk I'll describe recent st
udies focusing on reducing supervision for 3D scene understanding tasks, i
ncluding PointContrast -- a self-supervised representation learning framew
ork for 3D pointclods (ECCV 2020). Our findings in PointContrast are extre
mely encouraging: using a unified triplet of architecture, source dataset,
and contrastive loss for pre-training, we achieve improvement over recent
best results in segmentation and detection across 6 different benchmarks
for indoor and outdoor, real and synthetic datasets -- demonstrating that
the learned representation can generalize across domains.
Bio:
Or L
itany (PhD 2018, Tel-Aviv University) is a Research Scientist at NVIDIA's
Toronto AI lab, led by Prof. Sanja Fidler. Before that he was a postdoctor
al fellow at Stanford University, hosted by Prof. Leonidas Guibas, a postd
oc at Facebook AI Research, hosted by Prof. Jitendra Malik, and a postdoc
at the Technion, hosted by Prof. Alex Bronstein. Or received his B.Sc. in
Physics and Mathematics from the Hebrew University under the auspices of “
Talpiot”. He holds M.Sc. (Magna Cum Laude) and Ph.D. degrees in Electrical
Engineering from Tel-Aviv University (advised by Prof. Alex Bronstein). D
uring his PhD, Or has held visiting scholar appointments at TU Munich and
Duke universities and was a research intern at Microsoft Research, Intel,
and Google Research. Or's main interests include 3D deep learning, and met
hods for reducing supervision. He is the recipient of several awards inclu
ding a best paper award at SGP'16, best paper nomination at ICCV'19 and be
st paper award at ICML'20.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture:
91344952941
Meeting ID: 958 1720 7725
Passcode: CSLECTURE
UID:123se2401202435610
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20210118T110000
DTEND;TZID="Asia/Jerusalem":20210118T120000
DTSTAMP;TZID="Asia/Jerusalem":20210118T110000
FREEBUSY;FBTYPE=BUSY:20210118T110000/20210118T120000
SUMMARY;LANGUAGE=en-US:cggc talk by Myung Soo Kim (Seoul National Universi
ty) about CGGC Seminar: Accelerating Geometric Algorithms for Freeform Sur
faces using Toroidal Patch Approximation at 2021-01-18 11:00:00
DESCRIPTION;LANGUAGE=en-US:We present a new approach to the acceleration o
f geometric algorithms for freeform surfaces using a hierarchy of bounding
volumes, including those based on the osculating toroidal patches to the
surfaces. Using this approach, we revisit some non-trivial conventional ge
ometric algorithms, including those for computing the minimum and Hausdorf
f distances, the intersection and self-intersection curves, and the integr
al properties of freeform geometric models. We demonstrate the effectivene
ss of torus-based geometric computation, by reporting improvement in the s
peed, stability, and robustness of these algorithms.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture: 91344952941
UID:123se2401202435620
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20210119T103000
DTEND;TZID="Asia/Jerusalem":20210119T113000
DTSTAMP;TZID="Asia/Jerusalem":20210119T103000
FREEBUSY;FBTYPE=BUSY:20210119T103000/20210119T113000
SUMMARY;LANGUAGE=en-US:CSpecial Talk talk by Sarah Keren (Harvard Universi
ty and The Hebrew University of Jerusalem) about CS Lecture: Better Enviro
nments for Better AI at 2021-01-19 10:30:00
DESCRIPTION;LANGUAGE=en-US:Most AI research focuses exclusively on the AI
agent itself, i.e., given some input, what are the improvements to the age
nt’s reasoning that will yield the best possible output? In my research, I
take a novel approach to increasing the capabilities of AI agents via the
use of AI to design the environments in which they are intended to act. M
y methods identify the inherent capabilities and limitations of AI agents
and find the best way to modify their environment in order to maximize per
formance.
I will describe research projects that vary in their design
objectives, in the AI methodologies that are applied for finding optimal
designs, and in the real-world applications to which they correspond. One
example is Goal Recognition Design (GRD), which seeks to modify environmen
ts to allow an observing agent to infer the goals of acting agents as soon
as possible. A second is Helpful Information Shaping (HIS), which seeks
to find the minimal information to reveal to a partially-informed robot in
order to guarantee the robot’s goal can be achieved. I will also show how
HIS can be used in a market of information, where robots can trade their
knowledge about the environment and achieve an effective communication tha
t allows them to jointly maximize their performance. The third, Design for
Collaboration (DFC), considers an environment with multiple self-interest
ed reinforcement learning agents and seeks ways to encourage them to colla
borate effectively. Throughout the talk, I will discuss how the different
frameworks fit within my overarching objective of using AI to promote effe
ctive multi-agent collaboration and to enhance the way robots and machines
interact with humans.
Bio
Sarah Keren is a postdoctoral fellow at
The Harvard School of Engineering and Applied Sciences and The Hebrew Univ
ersity of Jerusalem. She received her PhD from the Technion, Israel Instit
ute of Technology. Sarah’s research focuses on providing theoretical found
ations for AI systems that are capable of effective collaboration with eac
h other and with people. She has received a number of awards, including th
e ICAPS 2020 Best Dissertation Honorable Mention, the ICAPS 2014 Honorable
Mention for Best Paper, the Eric and Wendy Schmidt Postdoctoral Award for
Women in Mathematical and Computing Sciences, and the Weizmann Institute
of Science National Postdoctoral Award for Advancing Women in Science.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture:
96384147559
Meeting ID: 963 8414 7559
Passcode: CSLECTURE
UID:123se2401202435640
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20210120T113000
DTEND;TZID="Asia/Jerusalem":20210120T123000
DTSTAMP;TZID="Asia/Jerusalem":20210120T113000
FREEBUSY;FBTYPE=BUSY:20210120T113000/20210120T123000
SUMMARY;LANGUAGE=en-US:ceClub talk by Julia Khamis (EE, Technion) about ce
Club: Demand-Aware Optimization in Offchain Networks at 2021-01-20 11:30:0
0
DESCRIPTION;LANGUAGE=en-US:Offchain networks are dominant as a solution to
the scalability problem of blockchain systems, allowing users to perform
payments without their recording on the chain by relying on predefined pay
ment channels. Users together with the offchain channels form a graph, kno
wn as the offchain network topology. A pair of users can employ a payment
even without a direct channel through a path of channels involving other i
ntermediate users. The offchain topology and payment characteristics affec
t network performance such as latency and fees.
We study basic demand
-aware problems in offchain network design: Efficiently mapping users to a
n offchain topology of a known structure as well as constructing a topolog
y of a bounded number of channels that can serve well typical payments. Li
kewise, we suggest an approach for jointly serving multiple payments by fi
nding an equivalent set of payments that has the same impact on user balan
ce but can be served efficiently in a given topology.
*MSc student u
nder supervision of Prof. Ori Rottenstreich
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture: for link to zoom please contact sgoren@campus.technion.ac.il
UID:123se2401202435630
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20210124T153000
DTEND;TZID="Asia/Jerusalem":20210124T163000
DTSTAMP;TZID="Asia/Jerusalem":20210124T153000
FREEBUSY;FBTYPE=BUSY:20210124T153000/20210124T163000
SUMMARY;LANGUAGE=en-US:msc talk by Ohad Goudsmid about Compositional Model
Checking for Multi-Properties at 2021-01-24 15:30:00
DESCRIPTION;LANGUAGE=en-US:Hyperproperties lift conventional trace propert
ies in a way that describes how a system behaves in its entirety, and not
just based on its individual traces.
We generalize this notion to mu
lti-properties, which describe the behavior of a set of systems, called a
multi-model. We show that model-checking multi-properties is equivalent to
model-checking hyperproperties.
We introduce sound and complete com
positional proof rules for model-checking multiproperties, based on approx
imations of the systems in the multi-model and describe methods of computi
ng them.
The first is abstraction-refinement based, in which a coars
e initial abstraction is continually refined using counterexamples, until
a suitable approximation is found.
The second, tailored for models w
ith finite traces, finds suitable approximations via the L* learning algor
ithm.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture: for link to zoom please contact goudsmidohad@cs.technion.ac.il
UID:123se2401202435500
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20210125T110000
DTEND;TZID="Asia/Jerusalem":20210125T120000
DTSTAMP;TZID="Asia/Jerusalem":20210125T110000
FREEBUSY;FBTYPE=BUSY:20210125T110000/20210125T120000
SUMMARY;LANGUAGE=en-US:cggc talk by Stefanie Elgeti (Institute of Lightwei
ght Design and Structural Biomechnics,TU Wien) about CGGC Seminar: Errors
in Judgement in Engineering: What Can They Teach Us about the Design Proce
ss? at 2021-01-25 11:00:00
DESCRIPTION;LANGUAGE=en-US:Engineering design is a task that comes with hi
gh responsibility: A failed design may easily cause not only monetary dama
ge but, even more importantly, injuries of users. Based on a collection of
design flaws [Petroski1994], this presentation will give an overview over
modern design approaches that can help to prevent these mistakes in the f
uture. It will touch upon both the topic of conceptual errors and numerica
l errors.
The lecture will not be recorded.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture: 91344952941
For password to lecture please contact inbalb@cs.technion.ac.il
UID:123se2401202435670
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20210127T113000
DTEND;TZID="Asia/Jerusalem":20210127T123000
DTSTAMP;TZID="Asia/Jerusalem":20210127T113000
FREEBUSY;FBTYPE=BUSY:20210127T113000/20210127T123000
SUMMARY;LANGUAGE=en-US:ceClub talk by Adam Morrison (Tel-Aviv University)
about ceClub: Comprehensive Protection for Speculatively-Accessed Data at
2021-01-27 11:30:00
DESCRIPTION;LANGUAGE=en-US:Speculative execution attacks present an enormo
us security threat, capable of reading arbitrary program data under malici
ous speculation and later exfiltrating that data over microarchitectural c
overt channels. This talk will describe a comprehensive hardware protectio
n from speculative execution attacks.
We will first describe Speculat
ive Taint Tracking (STT). STT delays the execution of instructions that cr
eate covert channels until their operands are proven to be a function of n
on-speculative data.
STT builds on a comprehensive characterization of c
overt channels on speculative microarchitectures and employs a novel micro
architecture for efficiently detecting when operands become non-speculativ
e and disabling protection at that time.
We will then describe Specul
ative Data-Oblivious Execution (SDO), which improves STT's performance by
executing covert-channel creating instructions in a data-oblivious manner,
i.e., so that their execution does not leak their operands. Data-obliviou
s execution usually implies doing the worst-case work all the time. SDO si
desteps this problem by using safe prediction to predict the work needed t
o satisfy the common case and subsequently perform it---all without leakin
g privacy.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture: for link to zoom please contact sgoren@campus.technion.ac.il
UID:123se2401202435650
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20210201T160000
DTEND;TZID="Asia/Jerusalem":20210201T170000
DTSTAMP;TZID="Asia/Jerusalem":20210201T160000
FREEBUSY;FBTYPE=BUSY:20210201T160000/20210201T170000
SUMMARY;LANGUAGE=en-US:CSpecial Talk talk by Yuval Moskovitch (University
of Michigan) about CS Lecture: Towards Reliable Data-Driven Computations a
t 2021-02-01 16:00:00
DESCRIPTION;LANGUAGE=en-US:Data-driven methods are increasingly being used
in domains such as fraud and risk detection, where data-driven algorithmi
c decision making may affect human life.
The growing impact of data a
nd data-driven systems on society makes it important that people be able t
o trust analytical results obtained from data-driven computations.
Th
is can be done in two complementary ways: by providing result explanations
so that the user understands the computation and the basis for the observ
ed results; and by profiling and monitoring the data used in the computati
on, to make the results more reliable in the first place.
In the firs
t part of the talk, I will present the use of provenance -- information re
garding the data origin and computational process -- for providing explana
tions of computational results. In the second part of the talk, I will pre
sent a method for data profiling using labels, as an example of a data-foc
used technique to facilitate an analyst building a reliable decision-makin
g pipeline.
Bio:
Yuval is a postdoctoral researcher at the Universi
ty of Michigan, hosted by Prof. H. V. Jagadish. Her research is centered a
round data management, advanced database applications, provenance, and pro
cess analysis. In her current research, she focuses on data management for
fairness and responsible data science. Yuval obtained her Ph.D. in Comput
er Science from Tel Aviv University, under the supervision of Prof. Daniel
Deutch. She completed a BSc in Software Engineering and MSc in Computer S
cience at Ben Gurion University. Yuval is the recipient of several awards
including the Shulamit Aloni Scholarship for Advancing Women in Science of
the Israeli Ministry of Science and Technology and the Data Science Fello
wship for outstanding postdocs of the planning and budgeting committee of
the council for higher education (VATAT).
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture:
97043323000
For password to lecture, please contact: sigal@cs.technion.ac.il
UID:123se2401202435700
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20210204T163000
DTEND;TZID="Asia/Jerusalem":20210204T173000
DTSTAMP;TZID="Asia/Jerusalem":20210204T163000
FREEBUSY;FBTYPE=BUSY:20210204T163000/20210204T173000
SUMMARY;LANGUAGE=en-US:CSpecial Talk talk by Hadar Averbuch-Elor (Cornell-
Tech) about CS Lecture: Deep into 3DV: Pushing the Boundaries of 3D Vision
at 2021-02-04 16:30:00
DESCRIPTION;LANGUAGE=en-US:3D computer vision has significantly advanced o
ver the past several decades, with modern algorithms successfully reconstr
ucting entire urban cities. However, many questions remain unexplored, as
geometric reasoning alone cannot fully infer the connections among images
capturing different parts of the scene or semantic relationships between i
mages captured at distant geographic locations.
In this talk, I will pr
esent an ongoing line of research that leverages powerful deep networks to
address new and exciting problems in 3D vision. Considering a single 3D s
cene, we ask: Can we estimate the relative camera rotation between a pair
of images in an extreme setting, where the images have little to no overla
p? We address this seemingly impossible task by designing a neural network
that can implicitly reason about hidden cues, such as vanishing points an
d direction of shadows. Expanding beyond a single scene, we jointly analyz
e dozens of 3D-augmented collections and connect them to a new domain: lan
guage. We demonstrate how a joint learned model that considers language, i
mages, and 3D geometry can reason about the rich semantics associated with
complex architectural landmarks. Finally, I will discuss several future d
irections.
Bio
Hadar Averbuch-Elor is a postdoctoral researcher at
Cornell-Tech working with Prof. Noah Snavely. Her research interests lie i
n the intersection of computer graphics and computer vision. Currently, he
r research focuses on understanding and manipulating visual concepts by co
mbining pixels with more structured modalities, including natural language
and 3D geometry. She completed her PhD in Electrical Engineering at Tel-A
viv University where she was advised by Prof. Daniel Cohen-Or. She has a B
.Sc. in Electrical Engineering from the Technion. She also held research p
ositions at Facebook and Amazon AI. Hadar is the recipient of several awar
ds including the Zuckerman Postdoctoral Scholar Fellowship and the Schmidt
Postdoctoral Award for Women in Mathematical and Computing Sciences.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture:
98635528430
For password to lecture, please contact: sigal@cs.technion.ac.il
UID:123se2401202435690
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20210207T170000
DTEND;TZID="Asia/Jerusalem":20210207T180000
DTSTAMP;TZID="Asia/Jerusalem":20210207T170000
FREEBUSY;FBTYPE=BUSY:20210207T170000/20210207T180000
SUMMARY;LANGUAGE=en-US:msc talk by Yotam Sharoni about Approximating Requi
rement Cut via a Configuration LP at 2021-02-07 17:00:00
DESCRIPTION;LANGUAGE=en-US:We consider the REQUIREMENT CUT problem, where
given an undirected graph G = (V, E) equipped with non-negative edge weigh
ts c , and g groups of vertices X1, . , Xg in V each equipped with a requi
rement ri, the goal is to find a collection of edges F in E, with total mi
nimum weight, such that once F is removed from G in the resulting graph ev
ery Xi is broken into at least ri connected components. REQUIREMENT CUT ca
ptures multiple classic cut problems in graphs, e.g., MULTICUT, MULTIWAY C
UT, MIN k-CUT, STEINER k-CUT, STEINER MULTICUT, and MULTI-MULTIWAY CUT. We
present an approximation for the problem, our algorithm is based on a new
configuration linear programming relaxation for the problem, which is acc
ompanied by a remarkably simple randomized rounding procedure.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture:
98726136846
For password to lecture, please contact: yotamsh@cs.technion.ac.il
UID:123se2401202435720
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20210208T100000
DTEND;TZID="Asia/Jerusalem":20210208T110000
DTSTAMP;TZID="Asia/Jerusalem":20210208T100000
FREEBUSY;FBTYPE=BUSY:20210208T100000/20210208T110000
SUMMARY;LANGUAGE=en-US:msc talk by Dor Hovav about Limited Associativity C
aching in the Data Plane at 2021-02-08 10:00:00
DESCRIPTION;LANGUAGE=en-US:In-network caching promises to improve the perf
ormance of distributed and networked applications.
This is by storing
so-called hot items in the network switches on-route between clients who
need access to the data and the storage servers who maintain it.
Sinc
e the data flows through those switches in any case, it is natural to cach
e hot items there.
Programmable switches enable managing such caches
in software, where the program gets compiled and then executed at ASIC spe
ed.
Yet, their limited programming model makes this task non-trivial.
Most software-managed caches treat the cache as a fully associative
region.
Alas, a fully associative design seems to be at odds with pro
grammable switches' goal of handling packets in a short-bounded amount of
time, as well as their restricted programming model.
Recently, the be
nefits of applying limited associative designs to software caches were stu
died and demonstrated.
In this work, we present PKache, a generic lim
ited associativity cache implementation in the programmable switches' doma
in-specific P4 language and demonstrate its utility by realizing multiple
popular cache management schemes.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture:
96832108498
For Password to lecture, please contact: dorhovav@cs.technion.ac.il
UID:123se2401202435660
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20210210T123000
DTEND;TZID="Asia/Jerusalem":20210210T133000
DTSTAMP;TZID="Asia/Jerusalem":20210210T123000
FREEBUSY;FBTYPE=BUSY:20210210T123000/20210210T133000
SUMMARY;LANGUAGE=en-US:Theory of Data talk by Dimitrios Myrisiotis (Comput
ing of Imperial College London) about Data Science & Deep Learning: One-ta
pe Turing Machine and Branching Program Lower Bounds for MCSP at 2021-02-1
0 12:30:00
DESCRIPTION;LANGUAGE=en-US:For a size parameter s: N -> N, the Minimum
Circuit Size Problem (denoted by
MCSP[s(n)]) is the problem of deciding
whether the minimum circuit size of a
given function f: {0,1}^n ->
{0,1} (represented by a string of length N := 2^n)
is at most a thresho
ld s(n). A recent line of work exhibited ``hardness
magnification'' phe
nomena for MCSP: A very weak lower bound for MCSP implies a
breakthroug
h result in complexity theory. For example, McKay, Murray, and
Williams
(STOC 2019) implicitly showed that, for some constant \mu_1 > 0, if
MCSP[2^{\mu_1 n}] cannot be computed by a one-tape Turing machine (with a
n
additional one-way read-only input tape) running in time N^{1.01}, th
en P != NP.
In this paper, we present the following new lower bounds
against one-tape Turing
machines and branching programs:
1. A ra
ndomized two-sided error one-tape Turing machine (with an additional
on
e-way read-only input tape) cannot compute MCSP[2^{\mu_2 n}] in time N^{1.
99},
for some constant \mu_2 > \mu_1.
2. A non-deterministic (
or parity) branching program of size o(N^{1.5} / \log N)
cannot compute
MKTP, which is a time-bounded Kolmogorov complexity analogue of
MCSP.
This is shown by directly applying the Neciporuk method to MKTP, which
previously appeared to be difficult.
3. The size of any non-determin
istic, co-non-deterministic, or parity branching
program computing MCSP
is at least N^{1.5-o(1)}.
These results are the first non-trivial l
ower bounds for MCSP and MKTP against
one-tape Turing machines and non-
deterministic branching programs, and
essentially match the best-known
lower bounds for any explicit functions against
these computational mod
els.
The first result is based on recent constructions of pseudorand
om generators for
read-once oblivious branching programs (ROBPs) and co
mbinatorial rectangles
(Forbes and Kelley, FOCS 2018; Viola 2019). En r
oute, we obtain several related
results:
1. There exists a (local
) hitting set generator with seed length
\tilde{O}(\sqrt{N}) secure aga
inst read-once polynomial-size non-deterministic
branching programs on
N-bit inputs.
2. Any read-once co-non-deterministic branching progra
m computing MCSP must have
size at least 2^{\tilde{\Omega}(N)}.
(
To appear STACS 2021).
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture:
96255595054
For password to lecture, please contact: mayasidis@cs.technion.ac.il
UID:123se2401202435710
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20210214T120000
DTEND;TZID="Asia/Jerusalem":20210214T130000
DTSTAMP;TZID="Asia/Jerusalem":20210214T120000
FREEBUSY;FBTYPE=BUSY:20210214T120000/20210214T130000
SUMMARY;LANGUAGE=en-US:phd talk by John Noonan about Indoor Exploration wi
th a Robotic Vehicle Using a Single Camera and a Floorplan at 2021-02-14 1
2:00:00
DESCRIPTION;LANGUAGE=en-US:Intelligent systems which can be deployed to ex
plore indoor buildings on a frequent and regular basis are beneficial to p
ersonnel operating remotely for security, manufacturing, or warehouse pack
-and-ship. In this talk, I will present a new minimalistic approach to in
door exploration: minimal sensing, minimal prior map knowledge, and minim
al underlying geometry needed to facilitate building a full visual scene r
epresentation. Our research combines both the classical and deep learning
worlds, harnessing the strengths of each, using a single camera and a floo
rplan to facilitate both indoor localization and building a full visual sc
ene representation of the explored building, with a small robotic vehicle
to carry out the exploration. We introduce a novel neural scene representa
tion that scales to full indoor buildings for view synthesis, describing i
t with a space of local neural rendering functions across the building whi
ch facilitates infusing meta-knowledge into the learning. Shared knowledge
of performing neural rendering from various vantage points in the scene i
s realized by conditioning on similar building structure, resulting in acc
elerated learning for the full building. We demonstrate learning such a ne
ural scene representation for view synthesis in around 15 minutes on a sin
gle commodity GPU and rendering in real-time at 64 Hz, allowing for immers
ive visual experiences.
Indoor exploration also requires accurate glo
bal positioning. We formulate a core methodology of integrating a floorpl
an with a monocular camera, forming the basis for our positioning systems
which resolve global position, orientation, and scale. We also present the
theoretical analysis of planar criteria for uniqueness of global localiza
tion solutions. We develop multiple algorithms to handle various necessary
components of indoor localization, such as extracting planes from scale-a
mbiguous monocular 3d pointclouds, associating extracted planes with floor
plan walls, recovering the scale factor from wall-plane pairs, and integra
ting soft vehicle and floorplan constraints in an optimization to refine g
lobal poses.
We introduce multiple modular global positioning systems
, both optimization-based and probabilistic approaches, and evaluate on cu
stom-created synthetic, simulation, and real-world datasets experimented u
sing a custom designed-and-built small robotic vehicle.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture:
2728213233
For password to lecture, please contact: John Noonan@cs.technion.ac.il
UID:123se2401202435750
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20210214T150000
DTEND;TZID="Asia/Jerusalem":20210214T160000
DTSTAMP;TZID="Asia/Jerusalem":20210214T150000
FREEBUSY;FBTYPE=BUSY:20210214T150000/20210214T160000
SUMMARY;LANGUAGE=en-US:phd talk by Alex Markuze about Characterizing, Expl
oiting, Detecting and Preventing DMA Attacks in the Presence of an IOMMU a
t 2021-02-14 15:00:00
DESCRIPTION;LANGUAGE=en-US:Malicious I/O devices might compromise the OS u
sing DMAs. The OS therefore utilizes the IOMMU to map and unmap every targ
et buffer right before and after its DMA is processed, thereby restricting
DMAs to their designated locations. This usage model, however, is neither
truly secure nor can it support multi-gigabit I/O operations.
IOMMU
provides protection at page granularity only, whereas DMA buffers can resi
de on the same page as other data leading to subpage vulnerabilities, whic
h make the system vulnerable to DMA attacks, in which I/O devices access a
nd manipulate memory regions not intended for their use. We first categori
ze subpage vulnerabilities into four categories, providing insight into th
e structure of DMA vulnerabilities. Then, to exploit these vulnerabilities
, we identify a set of three vulnerability attributes that are sufficient
to execute code injection attacks.
We then build analysis tools that
detect subpage vulnerabilities and analyze the Linux kernel. We find that
72% of the device drivers expose sensitive callback pointers, which may b
e overwritten by a device to hijack kernel control flow.
Aided by the
tools' output, we demonstrate novel code injection attacks on the Linux k
ernel we term Compound attacks. Specifically, while all previously reporte
d attacks are single-step, i.e., with the vulnerability attributes present
in a single page, in Compound attacks, the vulnerability attributes are i
nitially incomplete. However, they can be attained by carefully exploiting
standard OS behavior.
In order to provide performant and secure I/O
we propose that OSes utilize the IOMMU differently. Our new usage model re
stricts device access to a set of shadow DMA buffers that are never unmapp
ed. The DMAed data is copied to/from these shadow buffers, thus providing
sub-page protection. Our key insight is that the cost of interacting with,
and synchronizing access to the slow IOMMU hardware---required for zero-c
opy protection against devices---make copying preferable to zero-copying.
We implement our model in Linux and evaluate it with standard network
ing benchmarks utilizing a 40,Gb/s NIC. We demonstrate that despite being
more secure than the safest preexisting usage model, our approach provides
up to 5 times higher throughput. Additionally, whereas it is inherently l
ess scalable than an IOMMU-less (unprotected) system, our approach incurs
only 0%--25% performance degradation in comparison.
Next, we observe
that achieving protection at the DMA (un)map boundary is needlessly constr
aining, as devices must be prevented from changing the data only after the
kernel reads it. So there is no real need to switch ownership of buffers
between kernel and device at the DMA (un)mapping layer, as opposed to the
approach taken by all existing IOMMU protection schemes. We thus eliminate
the extra copy by (1)implementing a new allocator called DMA-Aware Malloc
for
Networking (DAMN), which (de)allocates packet buffers from a memory
pool permanently mapped in the IOMMU; (2)modifying the network stack to u
se this allocator; and (3)copying packet data only when the kernel needs i
t, which usually morphs the aforementioned extra copy into the kernel's st
andard copy operation performed at the user-kernel boundary. DAMN thus pro
vides full IOMMU protection with performance comparable to that of an unpr
otected system.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture:
99638464465
For password to lecture, please contact: markuze@cs.technion.ac.il
UID:123se2401202435760
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20210215T133000
DTEND;TZID="Asia/Jerusalem":20210215T143000
DTSTAMP;TZID="Asia/Jerusalem":20210215T133000
FREEBUSY;FBTYPE=BUSY:20210215T133000/20210215T143000
SUMMARY;LANGUAGE=en-US:msc talk by Ramy Masalha about Heterogeneous Parame
tric Trivariate Fillets at 2021-02-15 13:30:00
DESCRIPTION;LANGUAGE=en-US:Blending and filleting are well established ope
rations in solid modeling and computer-aided geometric design. The creatio
n of a transition surface which smoothly connects the boundary surfaces of
two (or more) objects has been extensively investigated. In this talk, we
will introduce several algorithms for the construction of, possibly heter
ogeneous, trivariate fillets, that support smooth filleting operations bet
ween pairs of, possibly heterogeneous, input trivariates.
A volumetri
c fillet, consisting of one or more tensor product trivariate(s), is fitte
d to the boundary surfaces of the input. The result smoothly blends betwee
n the two inputs, both geometrically and material-wise (properties of arbi
trary dimension).
Examples of all proposed algorithm will be also pre
sented.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture:
6222766056
For password to lecture, please contact: sramy@cs.technion.ac.il
UID:123se2401202435680
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20210216T113000
DTEND;TZID="Asia/Jerusalem":20210216T123000
DTSTAMP;TZID="Asia/Jerusalem":20210216T113000
FREEBUSY;FBTYPE=BUSY:20210216T113000/20210216T123000
SUMMARY;LANGUAGE=en-US:pixel-club talk by Marina Alterman (EE, Technion) a
bout Pixel Club: Imaging with Local Speckle Intensity Correlations: Theory
And Practice at 2021-02-16 11:30:00
DESCRIPTION;LANGUAGE=en-US:Recent advances in computational imaging have s
ignificantly expanded our ability to image through scattering layers such
as biological tissues, by exploiting the auto-correlation properties of ca
ptured speckle patterns. However, most experimental demonstrations of this
capability focus on the far-field imaging setting, where obscured light s
ources are very far from the scattering layer. By contrast, medical imagin
g applications such as fluorescent imaging operate in the near-field imagi
ng setting, where sources are inside the scattering layer. We provide a th
eoretical and experimental study of the similarities and differences betwe
en the two settings, highlighting the increased challenges posed by the ne
ar-field setting. We then draw insights from this analysis to develop a ne
w algorithm for imaging through scattering that is tailored to the near-f
ield setting, by taking advantage of unique properties of speckle patterns
formed under this setting, such as their local support. We present a theo
retical analysis of the advantages of our algorithm, and perform real expe
riments in both far-field and near-field configurations, showing an order-
of magnitude expansion in both the range and the density of the obscured p
atterns that can be recovered.
Short bio:
Marina Alterman received
the B.Sc. M.Sc. and Ph.D. degrees in electrical engineering from the Techn
ion under the supervision of Prof. Yoav Schechner. She was a Postdoctoral
Fellow in the Department of Electrical Engineering and Computer Science, N
orthwestern university, Evanston, IL, USA, during 2015–2017. She is curren
tly a researcher in the Computational Imaging lab (prof. Anat Levin) in th
e Electrical Engineering at the Technion.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture: https://technion.zoom.us/j/91594351204
UID:123se2401202435810
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20210217T113000
DTEND;TZID="Asia/Jerusalem":20210217T123000
DTSTAMP;TZID="Asia/Jerusalem":20210217T113000
FREEBUSY;FBTYPE=BUSY:20210217T113000/20210217T123000
SUMMARY;LANGUAGE=en-US:ceClub talk by Ori Lahav (Tel-Aviv University) abou
t ceClub: Designing a Programming Language Shared-Memory Concurrency Seman
tics at 2021-02-17 11:30:00
DESCRIPTION;LANGUAGE=en-US:A concurrency semantics (aka a memory model) fo
r a programming language defines the allowed behaviors of multithreaded pr
ograms. For programmers, sequential consistency (i.e., standard interleavi
ng-based semantics) is considered as the most intuitive model. However, it
is too costly to implement. Designing a satisfactory substitute is highly
challenging as it requires to carefully balance the conflicting desires o
f programmers, compilers, and hardware. In this talk I will introduce this
challenge and the key ideas behind the prototype example of the C/C++ con
currency model from 2011. Then, I will demonstrate the drawback of the C/C
++ approach, notoriously known as the "out-of-thin-air" problem. I will co
nclude by describing the "promising semantics" solution, and remaining cha
llenges.
Short bio:
Ori is a faculty member in the School of Comput
er Science at Tel Aviv University. He did his PhD at Tel Aviv University u
nder the supervision of Arnon Avron in the field of logic for computer sci
ence. In 2014, he was a postdoctoral researcher at Tel Aviv University hos
ted by Mooly Sagiv. After that, until September 2017, he was a postdoctora
l researcher at MPI-SWS in Germany hosted by Viktor Vafeiadis and Derek Dr
eyer. Ori's current main areas of research are programming languages and v
erification. Recently, Ori was awarded an ERC Starting Grant for the proje
ct titled "Verification-Aware Programming Language Concurrency Semantics".
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture:
93991018429
UID:123se2401202435790
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20210218T140000
DTEND;TZID="Asia/Jerusalem":20210218T150000
DTSTAMP;TZID="Asia/Jerusalem":20210218T140000
FREEBUSY;FBTYPE=BUSY:20210218T140000/20210218T150000
SUMMARY;LANGUAGE=en-US:CSpecial Event about Amazon Research, Alexa Shoppin
g Internship Program Introduction at 2021-02-18 14:00:00
DESCRIPTION;LANGUAGE=en-US:Amazon Research, Alexa Shopping Internship Prog
ram
Introduction on research challenges, and 2021 research
internshi
p program for graduate students in CS will be held on Thursday, Feb
ruary 18th between 14:00-15:00.
Agenda:
14:00 - 14:20 “Alexa c
an you help me shop?“ Yoelle Maarek, VP of Research, Alexa Shopping, Amazo
n
14:20 - 14:30 Introduction to the 2021 internship program, Liane L
ewin-Eytan, Sr Mgr., Alexa Shopping, Amazon
14:30 - 15:00 Panel &
; Q&As session, Moderated by Iftah Gamzu, Science Mgr, Alexa Shopping,
Amazon.
No Prior Knowledge is required but parti
cipation is by registration only and restricted to gradu
ate students and faculty members.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Event: Registration
UID:123se24012024100040
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20210218T143000
DTEND;TZID="Asia/Jerusalem":20210218T153000
DTSTAMP;TZID="Asia/Jerusalem":20210218T143000
FREEBUSY;FBTYPE=BUSY:20210218T143000/20210218T153000
SUMMARY;LANGUAGE=en-US:msc talk by Shahar Romem Peled about Batched Vertex
Cover Reconfiguration at 2021-02-18 14:30:00
DESCRIPTION;LANGUAGE=en-US:Our research focuses on the task of Batched Ver
tex Cover Reconfiguration, both in centralized and distributed systems. In
this talk, I will present a centralized black-box compression scheme for
reconfiguration schedules. Afterwards, I will introduce the concept of Sma
ll Separator Decomposition which can be used to compute schedules in distr
ibuted systems and show how to compute it on specific graph classes in the
LOCAL model of distributed computing. Lastly, I will complement the distr
ibuted results with a lower bound that shows that such decompositions are
somewhat necessary.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture:
99681314877
For password to lecture, please contact: shaharr@cs.technion.ac.il
UID:123se2401202435740
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20210223T113000
DTEND;TZID="Asia/Jerusalem":20210223T123000
DTSTAMP;TZID="Asia/Jerusalem":20210223T113000
FREEBUSY;FBTYPE=BUSY:20210223T113000/20210223T123000
SUMMARY;LANGUAGE=en-US:pixel-club talk by Amit Alfassy (EE, Technion) abou
t Pixel Club: Learning like Humans Do, with Limited Training Data at 2021
-02-23 11:30:00
DESCRIPTION;LANGUAGE=en-US:While Deep learning has brought a huge advancem
ent to computer vision, for most tasks we still need hundreds of labeled s
amples per class. The few-shot learning tasks attempts to alleviate the da
ta problem by learning from 1/ 5 samples per class. We will discuss the fe
w-shot learning domain through two of my papers. The first paper LaSO, is
a SOTA augmentation mechanic for multi-label few-shot classification and w
as published in CVPR 2019. The second paper StarNet is the SOTA weakly-sup
ervised few-shot object localization and detection method and was presente
d in AAAI 2021.
Short bio
Amit is a direct Ph.D. candidate at the
Electrical engineering faculty under the supervision of prof. Alex Bronste
in from the CS faculty. Amit also works part-time at IBM research AI.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture: https://technion.zoom.us/j/95741652165
UID:123se24012024100050
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20210225T110000
DTEND;TZID="Asia/Jerusalem":20210225T120000
DTSTAMP;TZID="Asia/Jerusalem":20210225T110000
FREEBUSY;FBTYPE=BUSY:20210225T110000/20210225T120000
SUMMARY;LANGUAGE=en-US:msc talk by Saar Eliad about Scalable deep learning
with pipeline model parallelism at 2021-02-25 11:00:00
DESCRIPTION;LANGUAGE=en-US:We worked on a particular case of Deep Learning
where the model is too large to fit into the memory of a single commodity
GPU during training. Such is the case for fine-tuning, an increasingly co
mmon technique that leverages transfer learning to dramatically expedite t
he training of huge, high-quality models. Critically, it holds the potenti
al to make giant state-of-the-art models pre-trained on high-end super-com
puting-grade systems readily available for users that lack access to such
costly resources.
In this seminar, we will present FTPipe, a system t
hat explores a previously unexplored dimension of pipeline model paralleli
sm, making multi-GPU execution of fine-tuning tasks for giant neural netwo
rks readily accessible. Our system goes beyond topology limitations of pre
vious pipeline-parallel approaches, efficiently training a new family of m
odels, including the current state-of-the-art. FTPipe achieves up to 3x sp
eedup and state-of-the-art accuracy when fine-tuning giant transformers wi
th billions of parameters.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture:
94960294313
For password to lecture, please contact: saareliad@cs.technion.ac.il
UID:123se24012024100020
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20210225T143000
DTEND;TZID="Asia/Jerusalem":20210225T153000
DTSTAMP;TZID="Asia/Jerusalem":20210225T143000
FREEBUSY;FBTYPE=BUSY:20210225T143000/20210225T153000
SUMMARY;LANGUAGE=en-US:msc talk by Noa Marelly about Fault Tolerant Max-Cu
t at 2021-02-25 14:30:00
DESCRIPTION;LANGUAGE=en-US:In this work, we initiate the study of fault to
lerant Max-Cut, where given an edge-weighted undirected graph G=(V,E), the
goal is to find a cut S, that maximizes the total weight of edges that cr
oss S even after an adversary removes k vertices from G.
We consider
two types of adversaries: an adaptive adversary that sees the outcome of t
he random coin tosses used by the algorithm, and an oblivious adversary th
at does not.
For any constant number of failures k we present an appro
ximation of 0.878 against an adaptive adversary and of 0.8786 against an o
blivious adversary (here 0.8786 is the approximation achieved by the rando
m hyperplane algorithm of [Goemans-Williamson J. ACM `95]).
Additional
ly, we present a hardness of approximation of 0.8786 against both types of
adversaries, rendering our results (virtually) tight.
Based on joint w
ork with Keren Censor-Hillel, Roy Schwartz and Tigran Tonoyan.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture:
98844121807
For password to lecture, please contact: noa.marelly@cs.technion.ac.il
UID:123se2401202435800
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20210225T193000
DTEND;TZID="Asia/Jerusalem":20210225T233000
DTSTAMP;TZID="Asia/Jerusalem":20210225T193000
FREEBUSY;FBTYPE=BUSY:20210225T193000/20210225T233000
SUMMARY;LANGUAGE=en-US:CSpecial Event about Google Hash Code 2021 at 2021-
02-25 19:30:00
DESCRIPTION;LANGUAGE=en-US:Google Hash Code 2021will take place on Thursda
y, February 25, 2021 between 19:30-23:45 and you are invited to register t
o the Technion Hub by Wednesday, February 24, 13:00 IST.
More det
ails and registration.
Technion Hub Facebook Group
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Event: Registration
UID:123se24012024100030
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20210302T113000
DTEND;TZID="Asia/Jerusalem":20210302T123000
DTSTAMP;TZID="Asia/Jerusalem":20210302T113000
FREEBUSY;FBTYPE=BUSY:20210302T113000/20210302T123000
SUMMARY;LANGUAGE=en-US:phd talk by Chaim Baskin about Designing Deep Neura
l Networks for Efficient and Robust Inference at 2021-03-02 11:30:00
DESCRIPTION;LANGUAGE=en-US:Deep neural networks (DNN) became a common tool
for solving complex tasks in various fields such as computer vision, natu
ral language processing, and recommendation systems. Despite recent progre
ss made in enhancing the DNN performance, there are still two major obstac
les hindering the practicality of DNNs in some application: their energy-e
xpensive deployment on embedded platforms, and their amenability to malici
ous adversarial perturbations. In this talk, I will overview several lines
of works tackling different aspects of both problems. The first presents
two training-aware and post-training quantization approaches making the DN
Ns parameters and feature maps represented in fixed low-bit representation
. The second introduces two entropy coding-based methods for the reduction
of inference-time memory bandwidth requirements; the first method does no
t require any fine-tuning, while the second includes a fine-tuning stage a
nd in exchange provides significant further bandwidth reduction with negli
gible additional complexity or accuracy reduction. I will also present a s
imple framework that helps design efficient hardware for quantized neural
networks. I will show how quantization techniques can inspire new approach
es to better coping with adversarial attacks and demonstrate how an advers
arially pre-trained classifier could boost adversarial robustness by smoot
hing between different levels of input noise. Finally, I will introduce a
simple single-node minimal attribute changing perturbation that can attack
social graph-based DNNs, in a significantly more harmful way than the pre
viously studied edge-based attacks.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture:
99572398109
For password to lecture, please contact: chaimbaskin@cs.technion.ac.il
UID:123se2401202435730
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20210302T170000
DTEND;TZID="Asia/Jerusalem":20210302T180000
DTSTAMP;TZID="Asia/Jerusalem":20210302T170000
FREEBUSY;FBTYPE=BUSY:20210302T170000/20210302T180000
SUMMARY;LANGUAGE=en-US:phd talk by Hadar Frenkel about Automata over Infin
ite Data Domains: Learnability and Applications in Program Verification an
d Repair at 2021-03-02 17:00:00
DESCRIPTION;LANGUAGE=en-US:We present automata over infinite data domains
and their use in program verification and repair.
In particular, we d
iscuss assume-guarantee based verification, a compositional verification m
ethod that uses automata learning in order to modularly verify the correct
ness of a system.
Then we present Assume-Guarantee-Repair (AGR) – a f
ramework that verifies that a program satisfies a set of properties, and r
epairs the program in case the verification fails. We consider communicati
ng programs – these are simple C-like programs, extended with synchronous
communication actions over communication channels.
We model these inf
inite-state systems using finite automata, and reduce the semantic problem
s in hand – satisfying complex specifications that also contain first-orde
r constraints – to syntactic ones, namely membership and equivalence queri
es for regular languages.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture:
97090529670
For password to lecture, please contact: hfrenkel@cs.technion.ac.il
UID:123se2401202435780
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20210304T123000
DTEND;TZID="Asia/Jerusalem":20210304T133000
DTSTAMP;TZID="Asia/Jerusalem":20210304T123000
FREEBUSY;FBTYPE=BUSY:20210304T123000/20210304T133000
SUMMARY;LANGUAGE=en-US:msc talk by Noam Yefet about Adversarial Examples f
or Models of Code and Defending Against Them at 2021-03-04 12:30:00
DESCRIPTION;LANGUAGE=en-US:Neural models of code have shown impressive res
ults when performing tasks such as
predicting method names and identifyi
ng certain kinds of bugs. We show that these
models are vulnerable to ad
versarial examples, and introduce a novel approach for
attacking trained
models of code using adversarial examples. The main idea of our
approac
h is to force a given trained model to make an incorrect prediction, as sp
ecified
by the adversary, by introducing small perturbations that do not
change the program’s
semantics, thereby creating an adversarial example
. To find such perturbations, we
present a new technique for Discrete Ad
versarial Manipulation of Programs (DAMP).
DAMP works by deriving the de
sired prediction with respect to the model’s inputs,
while holding the m
odel weights constant, and following the gradients to slightly modify
th
e input code.
We show that our DAMP attack is effective across three
neural architectures:
code2vec, GGNN, and GNN-FiLM, in both Java and C#.
Our evaluations demonstrate
that DAMP has up to 89% success rate in cha
nging a prediction to the adversary’s
choice (a targeted attack) and a s
uccess rate of up to 94% in changing a given prediction
to any incorrect
prediction (a non-targeted attack).
To defend a model against such a
ttacks, we empirically examine a variety of possible defenses
and discus
s their trade-offs. We show that some of these defenses can dramatically d
rop the
success rate of the attacker, with a minor penalty of 2% relativ
e degradation in accuracy
when they are not performing under attack.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture:
96898381897
For password to lecture, please contact: snyefet@cs.technion.ac.il
UID:123se24012024100070
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20210309T110000
DTEND;TZID="Asia/Jerusalem":20210309T120000
DTSTAMP;TZID="Asia/Jerusalem":20210309T110000
FREEBUSY;FBTYPE=BUSY:20210309T110000/20210309T120000
SUMMARY;LANGUAGE=en-US:msc talk by Michael Ezra about Small Circuits Imply
Efficient Arthur-Merlin Protocols at 2021-03-09 11:00:00
DESCRIPTION;LANGUAGE=en-US:We show a new connection between circuit lower
bounds and interactive proofs in restricted computational models.
Spe
cifically, we focus on the frontier problem of whether a DNF augmented wit
h an additional layer of parity (XOR) gates, can approximate the inner pro
duct function.
We show that the existence of such a small circuit, wo
uld have unexpected general implications for interactive variants of the D
ata Streaming and Communication Complexity models.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture:
5617822865
For password to lecture, please contact: michaelezra@cs.technion.ac.il
UID:123se24012024100110
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20210309T113000
DTEND;TZID="Asia/Jerusalem":20210309T123000
DTSTAMP;TZID="Asia/Jerusalem":20210309T113000
FREEBUSY;FBTYPE=BUSY:20210309T113000/20210309T123000
SUMMARY;LANGUAGE=en-US:pixel-club talk by Michael Bronstein (Imperial Coll
ege London) about Pixel Club: Geometric Deep Learning: the Erlangen Progra
mme of ML at 2021-03-09 11:30:00
DESCRIPTION;LANGUAGE=en-US:For nearly two millennia, the word "geometry" w
as synonymous with Euclidean geometry, as no other types of geometry exist
ed. Euclid's monopoly came to an end in the 19th century, where multiple e
xamples of non-Euclidean geometries were shown. However, these studies qui
ckly diverged into disparate fields, with mathematicians debating the rela
tions between different geometries and what defines one. A way out of this
pickle was shown by Felix Klein in his Erlangen Programme, which proposed
approaching geometry as the study of invariants or symmetries using the l
anguage of group theory. In the 20th century, these ideas have been fundam
ental in developing the modern physics, culminating in the Standard Model.
The current state of deep learning somewhat resembles the situation
in the field of geometry in the 19h century: On the one hand, in the past
decade deep learning has brought a revolution in data science and made po
ssible many tasks previously thought to be beyond reach -- including compu
ter vision, playing Go, or protein folding. At the same time, we have a zo
o of neural network architectures for various kinds of data, but few unify
ing principles. As in times past, it is difficult to understand the relati
ons between different methods, inevitably resulting in the reinvention and
re-branding of the same concepts.
Geometric Deep Learning aims to
bring geometric unification to deep learning in the spirit of the Erlangen
Programme. Such an endeavour serves a dual purpose: it provides a common
mathematical framework to study the most successful neural network archite
ctures, such as CNNs, RNNs, GNNs, and Transformers, and gives a constructi
ve procedure to incorporate prior knowledge into neural networks and build
future architectures in a principled way. In this talk, I will overview t
he mathematical principles underlying Geometric Deep Learning on grid, gra
phs, and manifolds, and show some of the exciting and groundbreaking appli
cations of these methods in the domains of computer vision, social science
, biology, and drug design.
(based on joint work with J. Bruna, T. C
ohen, P. Veličković)
Bio:
Michael Bronstein is a professor at Imper
ial College London, where he holds the Chair in Machine Learning and Patte
rn Recognition, and Head of Graph Learning Research at Twitter. He also he
ads ML research in Project CETI, a TED Audacious Prize-winning collaborati
on aimed at understanding the communication of sperm whales. Michael recei
ved his PhD from the Technion in 2007. He has held visiting appointments a
t Stanford, MIT, Harvard, and TAU, and has also been affiliated with three
Institutes for Advanced Study (at TUM as a Rudolf Diesel Fellow (2017-201
9), at Harvard as a Radcliffe fellow (2017-2018), and at Princeton as a vi
sitor (2020)). Michael is the recipient of the Royal Society Wolfson Resea
rch Merit Award, Royal Academy of Engineering Silver Medal, five ERC grant
s, two Google Faculty Research Awards, and two Amazon AWS ML Research Awar
ds. He is a Member of the Academia Europaea, Fellow of IEEE, IAPR, BCS, an
d ELLIS, ACM Distinguished Speaker, and a World Economic Forum Young Scien
tist. In addition to his academic career, Michael is a serial entrepreneur
and founder of multiple startup companies, including Novafora, Invision (
acquired by Intel in 2012), Videocites, and Fabula AI (acquired by Twitter
in 2019). He has previously served as Principal Engineer at Intel Percept
ual Computing and was one of the developers of the Intel RealSense technol
ogy.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture: https://technion.zoom.us/j/94556114100
UID:123se24012024100090
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20210309T140000
DTEND;TZID="Asia/Jerusalem":20210309T150000
DTSTAMP;TZID="Asia/Jerusalem":20210309T140000
FREEBUSY;FBTYPE=BUSY:20210309T140000/20210309T150000
SUMMARY;LANGUAGE=en-US:msc talk by Or Goaz about Clustering in the Network
Data Plane at 2021-03-09 14:00:00
DESCRIPTION;LANGUAGE=en-US:Clustering is a basic machine learning task. In
this task, a stream of input items needs to be grouped into clusters, suc
h that all items classified into the same cluster are closer to each other
than to items classified to other clusters. Each cluster is centered arou
nd a centroid point, which may either be given as a parameter, or must be
learned during the process in the case of unsupervised online learning.
This work studies the ability to perform clustering in programmable swi
tches. The motivation for using programmable switches comes from the fact
that classifying network traffic is a basic need for improved network secu
rity and management. Conducting such classification by the switches throug
h which the traffic flows is potentially the most efficient approach. To t
hat end, we develop Clustreams, a novel in-network clustering system desig
ned to handle clustering in the data path. At the core of Clustreams is a
novel clustering algorithm that relies heavily on TCAM (Ternary Content Ad
dressable Memory) match-action capabilities. This algorithm is realized fo
r the Nvidia Spectrum-3 switch, and is limited to classification when the
centroid points are known a-priori. We also present an extension of Clustr
eams that supports unsupervised online learning.
The work includes ac
curacy measurements for the algorithms, as well as run-time performance me
asurements and analysis of the clustering algorithm on a Spectrum-3 switch
. As shown in the measurements, Clustreams obtains very high accuracy with
negligible run-time overheads.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture:
99911513639
For password to lecture, please contact: orgoaz@cs.technion.ac.il
UID:123se24012024100060
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20210309T170000
DTEND;TZID="Asia/Jerusalem":20210309T190000
DTSTAMP;TZID="Asia/Jerusalem":20210309T170000
FREEBUSY;FBTYPE=BUSY:20210309T170000/20210309T190000
SUMMARY;LANGUAGE=en-US:CSpecial Event about Verizon Media Internship Meetu
p at 2021-03-09 17:00:00
DESCRIPTION;LANGUAGE=en-US:CS graduate studies students are invited to an
Internship Meetup by Verizon Media, on Tuesday, Mach 9, 2021, 17:00.
For participation please pre-register by email.
More details
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Event: Registration
UID:123se24012024100140
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20210310T113000
DTEND;TZID="Asia/Jerusalem":20210310T123000
DTSTAMP;TZID="Asia/Jerusalem":20210310T113000
FREEBUSY;FBTYPE=BUSY:20210310T113000/20210310T123000
SUMMARY;LANGUAGE=en-US:ceClub talk by Eyal Ronen (Tel-Aviv University) abo
ut ceClub: Dragonblood: Analyzing the Dragonfly Handshake of WPA3 and EAP-
pwd at 2021-03-10 11:30:00
DESCRIPTION;LANGUAGE=en-US:The WPA3 certification aims to secure home netw
orks, while EAP-pwd is used by certain enterprise WiFi networks to authent
icate users. Both use the Dragonfly handshake to provide forward secrecy a
nd resistance to dictionary attacks. In this paper, we systematically eval
uate Dragonfly's security. First, we audit implementations, and present ti
ming leaks and authentication bypasses in EAP-pwd and WPA3 daemons. We the
n study Dragonfly's design and discuss downgrade and denial-of-service att
acks. Our next and main results are side-channel attacks against Dragonfly
's password encoding method (e.g.~hash-to-curve).
We believe that the
se side-channel leaks are inherent to Dragonfly. For example, after our in
itial disclosure, patched software was still affected by a novel side-chan
nel leak. We also analyze the complexity of using the leaked information t
o brute-force the password. For instance, brute-forcing a dictionary of si
ze 10^10 requires less than $1 in Amazon EC2 instances. These results are
also of general interest due to ongoing standardization efforts on Dragonf
ly as a TLS handshake, Password-Authenticated Key Exchanges (PAKEs), and h
ash-to-curve. Finally, we discuss backwards-compatible defenses, and propo
se protocol fixes that prevent attacks. Our work resulted in a new draft o
f the protocols incorporating our proposed design changes.
*Joint wor
k by Mathy Vanhoef (New York University Abu Dhabi) and Eyal Ronen(TAU) .
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture:
94193733696
UID:123se24012024100120
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20210310T163000
DTEND;TZID="Asia/Jerusalem":20210310T173000
DTSTAMP;TZID="Asia/Jerusalem":20210310T163000
FREEBUSY;FBTYPE=BUSY:20210310T163000/20210310T173000
SUMMARY;LANGUAGE=en-US:phd talk by Idan Schwartz about Cognitive Models in
Deep Learning at 2021-03-10 16:30:00
DESCRIPTION;LANGUAGE=en-US:The quest for algorithms that enable cognitive
abilities is an integral part of machine learning and appears in many face
ts, such as virtual assistant and visual reasoning. A cognitive system req
uires an effective approach to extract details and nuances from the multip
le sensors that pound the devices' computational engine. To this end, we p
ropose a novel form of attention mechanism, namely Factor Graph Attention,
that operates on any data utilities and differentiates useful signals fro
m distracting ones.
Our model won the Visual Dialog challenge and sh
owed a state-of-the-art performance on various tasks, such as Visual Quest
ion Answering (VQA), Video Dialog, and Visual Storytelling. Despite the su
bstantial improvements the attention mechanism has been permitting, strong
classifiers are prone to exploit biases and find shortcuts. As a conseque
nce, current methods may solve the dataset but not directly the task. To a
ddress this concern, we introduce perceptual scores that assess the degree
to which a model relies on the input features' different subsets (i.e., m
odalities). We also propose methods to increase a model's perceptiveness,
such as sample re-weighting and an information-based regularization. We va
lidate our methods' efficacy on various datasets, such as VQA, SocialIQ,
and SNLI.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture:
9855273458
For password to lecture, please contact: idansc@cs.technion.ac.il
UID:123se24012024100100
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20210314T110000
DTEND;TZID="Asia/Jerusalem":20210314T120000
DTSTAMP;TZID="Asia/Jerusalem":20210314T110000
FREEBUSY;FBTYPE=BUSY:20210314T110000/20210314T120000
SUMMARY;LANGUAGE=en-US:msc talk by Asaf Yeshurun about Extracting Bible Qu
otes from Historical Commentary at 2021-03-14 11:00:00
DESCRIPTION;LANGUAGE=en-US:The Hebrew Bible (Tanach) has been extensively
quoted by historical religious text and commentaries throughout history.
Nowadays, many of these text resources are publicly available online.
Yet, the Bible quotations within them are often partially identified if a
t all.
Knowing the exact quotations may be highly beneficial to scho
lars interested in studying or investigating the Bible.
We have deve
loped and empirically analyzed a machine-learning solution for this task.
End-to-end, our model is comprised of three main stages: (a) rule-bas
ed candidate generation, (b) context extraction using available historical
commentary, and (c) an artificial neural-network for candidate scoring.
To evaluate our models, we have constructed labeled data based on the
Hebrew Bible commentary known as Midrash Raba, which contains more than h
alf a million words and over 30,000 quotations.
Our solution scores
over 80% F-score, and considerably outperforms several state-of-the-art ap
proaches for tasks of similar nature. As a contribution of independent int
erest, our solution includes of a novel word-embedding method that seeks t
o utilize the nature of our text and its context.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture:
5201760342
For password to lecture, please contact: asafyeshurun@cs.technion.ac.il
UID:123se24012024100130
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20210315T180000
DTEND;TZID="Asia/Jerusalem":20210315T190000
DTSTAMP;TZID="Asia/Jerusalem":20210315T180000
FREEBUSY;FBTYPE=BUSY:20210315T180000/20210315T190000
SUMMARY;LANGUAGE=en-US:msc talk by Aviv Ben-David about Investigating the
Difference Between Emulated and Paravirtual Network I/O: The Strange, Unto
ld Story at 2021-03-15 18:00:00
DESCRIPTION;LANGUAGE=en-US:In virtual setups, guest virtual machines (VMs)
perform their I/O through virtual I/O devices that are implemented by the
hypervisor in software. There are two major flavors of virtual I/O device
s. The first is ``emulation’’, which provides an interface identical to th
at of some preexisting physical I/O device, thus allowing the operating sy
stem (OS) inside the VM to use the original driver of the device, as is, u
naware that it is in fact virtual (implemented in software). The second so
ftware indirection layer flavor is ``paravirtualization’’, which makes the
VM aware that it is being virtualized. The purpose of paravirtual devices
is to leverage this awareness to reduce the number of virtualization exit
s (context switches between guest and hypervisor) and thus improve perform
ance.
Focusing on networking, we observe that the performance of emul
ation is so much worse than that of paravirtualization, that it significan
tly reduces the utility of emulation and effectively prevents performance-
sensitive users from enjoying the benefits. We hypothesize that the poor p
erformance of emulation is unjustified and can be rectified. Our hypothesi
s is based on a simple model that accounts for the additional virtualizati
on exits incurred by emulation relative to paravirtualization. The model i
ndicates that the poor performance exhibited by emulation could be due to
issues different than exits, suggesting that these issues may be resolvabl
e.
We systematically analyze the implementation of emulation as compa
red to paravirtualization. We uncover the problems underlying the performa
nce difference and find that, indeed, they can largely be resolved.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture:
96844553386
For password to lecture, please contact: bdaviv@cs.technion.ac.il
UID:123se24012024100230
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20210316T113000
DTEND;TZID="Asia/Jerusalem":20210316T123000
DTSTAMP;TZID="Asia/Jerusalem":20210316T113000
FREEBUSY;FBTYPE=BUSY:20210316T113000/20210316T123000
SUMMARY;LANGUAGE=en-US:msc talk by Amit Bracha about Shape correspondence
by aligning scale-invariant LBO eigenfunctions at 2021-03-16 11:30:00
DESCRIPTION;LANGUAGE=en-US:When matching non-rigid shapes, the regular or
scale-invariant Laplace-Beltrami Operator (LBO) eigenfunctions could poten
tially serve as intrinsic descriptors which are invariant to isometric tra
nsformations. However, the computed eigenfunctions of two quasi-isometric
surfaces could be substantially different. Such discrepancies include sign
ambiguities and possible rotations and reflections within subspaces spann
ed by eigenfunctions that correspond to similar eigenvalues. Thus, without
aligning the corresponding eigenspaces it is difficult to use the eigenfu
nctions as descriptors. In this talk, we will propose to model the relati
ve transformation between the eigenspaces of two quasi-isometric shapes us
ing a band orthogonal matrix, as well as present a framework that aims to
estimate this matrix. We will show that estimating this transformation all
ows us to align the eigenfunctions of one shape with those of the other, t
hat could then be used as intrinsic, consistent, and robust descriptors. T
o estimate the transformation we use an unsupervised spectral-net framewor
k that uses descriptors given by the eigenfunctions of the scale-invariant
version of the LBO. Then, using a spectral training mechanism, we find a
band limited orthogonal matrix that aligns the two sets of eigenfunctions.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture:
3615145651
For password to lecture, please contact: amitbracha@cs.technion.ac.il
UID:123se24012024100240
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20210317T143000
DTEND;TZID="Asia/Jerusalem":20210317T153000
DTSTAMP;TZID="Asia/Jerusalem":20210317T143000
FREEBUSY;FBTYPE=BUSY:20210317T143000/20210317T153000
SUMMARY;LANGUAGE=en-US:msc talk by Eitan Kosman about Complex Pattern Mini
ng at 2021-03-17 14:30:00
DESCRIPTION;LANGUAGE=en-US:Mining complex patterns from large data sets ha
s attracted much attention in the last few decades. A plethora of methods
and algorithms have been designed for mining a variety of patterns, rangin
g from simple association rules and frequent itemsets to advanced graph-ba
sed structures. However, as modern applications grow dramatically more sop
histicated and operate on highly multidimensional and increasingly complex
data, they introduce the demand for mining even more expressive and convo
luted patterns unsupported by the current state-of-the-art techniques. As
a result, more powerful and expressive pattern mining approaches are neede
d.
We propose a novel method for multi-item multi-attribute pattern m
ining (MIMA-PM) - a generalization of classic pattern mining to substantia
lly more expressive patterns. To the best of our knowledge, this work is t
he first to formally define and properly address this highly important pro
blem. Extensive experimental evaluation conducted on synthetic and real-wo
rld data demonstrates high accuracy and scalability of our method.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture:
96049971966
For password to lecture, please contact: eitan.k@cs.technion.ac.il
UID:123se24012024100160
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20210321T110000
DTEND;TZID="Asia/Jerusalem":20210321T120000
DTSTAMP;TZID="Asia/Jerusalem":20210321T110000
FREEBUSY;FBTYPE=BUSY:20210321T110000/20210321T120000
SUMMARY;LANGUAGE=en-US:msc talk by Moshe Sebag about The Shapley Value of
Tuples in Query Answering at 2021-03-21 11:00:00
DESCRIPTION;LANGUAGE=en-US:This research aims to investigate the applicati
on of the Shapley value to quantify the contribution of a tuple to a query
answer. The Shapley value is a widely known numerical measure in cooperat
ive game theory and in many applications of game theory for assessing the
contribution of a player to a coalition game. It has been established alre
ady in the 1950s, and is theoretically justified by being the very single
wealth distribution measure that satisfies some natural axioms. While this
value has been investigated in several areas, it received little attentio
n in data management. The aforementioned qualities of the Shapley value ma
ke it a better measure than several measures which have been suggested to
quantify the contribution of a tuple recently.
Furthermore, it allow
s us to compute responsibility with aggregation queries.
We study thi
s measure in the context of conjunctive and aggregate queries by defining
corresponding coalitional games. We establish a dichotomy in complexity fo
r the class of Boolean conjunctive queries without self-joins. In addition
, we provide an efficient algorithm to compute the Shapley value in the tr
actable cases; and for the hard cases we present approximation algorithm.
In the practical aspect, we implement these algorithms to study them
empirically and find strategies and techniques to optimize them towards a
practical application on realistic data.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture:
8029792183
For password to lecture, please contact: moshesebag@cs.technion.ac.il
UID:123se24012024100260
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20210321T170000
DTEND;TZID="Asia/Jerusalem":20210321T190000
DTSTAMP;TZID="Asia/Jerusalem":20210321T170000
FREEBUSY;FBTYPE=BUSY:20210321T170000/20210321T190000
SUMMARY;LANGUAGE=en-US:CSpecial Event about To Foresee the Future - The Pr
ediction that will Save the World - A Lecture by Dr. Kira Radinsky at 2021
-03-21 17:00:00
DESCRIPTION;LANGUAGE=en-US:You are invited to a lecture by Dr. Kira Radinsky: "To Fo
resee the Future - The Prediction that will Save the World", on Sunday, Ma
rch 21, 17:00.
A link to the Zoom meeting will be sent upon pre-registration.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Event: Registration
UID:123se24012024100210
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20210321T193000
DTEND;TZID="Asia/Jerusalem":20210321T203000
DTSTAMP;TZID="Asia/Jerusalem":20210321T193000
FREEBUSY;FBTYPE=BUSY:20210321T193000/20210321T203000
SUMMARY;LANGUAGE=en-US:CSpecial Event about A Meeting on Webinar by Huawei
at 2021-03-21 19:30:00
DESCRIPTION;LANGUAGE=en-US:CS graduate studies students are invited to a m
eeting on Webinar by Huawei, on Sunday, March 21, 2021, 19:30-20:30/
For participation please pre-register by email.
More details
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Event: Registration
UID:123se24012024100150
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20210322T110000
DTEND;TZID="Asia/Jerusalem":20210322T120000
DTSTAMP;TZID="Asia/Jerusalem":20210322T110000
FREEBUSY;FBTYPE=BUSY:20210322T110000/20210322T120000
SUMMARY;LANGUAGE=en-US:cggc talk by Stefanie Hahmann (University Grenoble
INP) about CGGC Seminar: Geometric Construction of Auxetic Metamaterials a
t 2021-03-22 11:00:00
DESCRIPTION;LANGUAGE=en-US:Recent advances in digital manufacturing, where
computational design, materials science and engineering meet, offer whole
new perspectives for tailoring mechanical properties and fabrication of m
aterial with applications as diverse as product design, architecture, engi
neering and art. Auxetic materials are characterized by a negative Poisson
’s ratio. This means that they do not behave as usual materials. When stre
tched in one direction, they do not shrink in the other directions, in con
trary they expand. In comparison to standard materials, auxetics are there
fore characterized by enhanced mechanical properties such as energy absorp
tion, indentation resistance and acoustic absorption.
This presentati
on is devoted to our recent work on a category of metamaterials called aux
etic structures, or auxetic networks. Whereas regular auxetic networks are
well studied, our focus is on irregular, also called disordered auxetic n
etworks. In particular, we are exploring geometrical strategies to generat
e 2-dimensional disordered auxetic structures.
Starting from an irre
gular dense network, we seek to reduce the Poisson's ratio, by pruning bon
ds (edges) based solely on geometric criteria. To this end, we first deduc
e some prominent geometric features from regular auxetic networks and then
introduce a strategy combining a pure geometric pruning algorithm followe
d by a physics-based testing phase to determine the resulting Poisson's ra
tio of our networks. We provide statistical validation of our approach on
large sets of irregular networks, and we additionally show real auxetic ne
tworks laser-cut using sheets of rubber. The findings reported here show t
hat it is possible to reduce the Poisson's ratio by geometric pruning, and
that we can generate disordered auxetic networks at lower processing time
s than a physics-based approach.
The lecture will not be recorded.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture: 91344952941
For password to lecture please contact inbalb@cs.technion.ac.il
UID:123se24012024100270
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20210324T160000
DTEND;TZID="Asia/Jerusalem":20210324T170000
DTSTAMP;TZID="Asia/Jerusalem":20210324T160000
FREEBUSY;FBTYPE=BUSY:20210324T160000/20210324T170000
SUMMARY;LANGUAGE=en-US:msc talk by Sagi Marcovich about Reconstruction of
Strings from their Substrings Spectrum at 2021-03-24 16:00:00
DESCRIPTION;LANGUAGE=en-US:Using DNA molecules as a data storage volume wa
s first introduced in the 1960s by Richard Feynman. Later, in 1990, the
human genome project led to a significant progress in sequencing and asse
mbly methods. As a result, the interest in storage solutions based on DNA
molecules was increased. DNA storage enjoys major advantages over magnetic
and optical storage solutions.
Motivated by rising technologies for
DNA sequencing, this work studies reconstruction of strings based upon the
ir substrings spectrum. Under this paradigm, it is assumed that all substr
ings of some fixed length are received and the goal is to reconstruct the
string. While many existing works assumed that substrings are received err
or free, we follow in this paper the noisy setup of this problem that was
first studied by Gabrys and Milenkovic. The goal of this study is twofold.
First we study the setup in which not all substrings in the multispectrum
are received, and then we focus on the case where the read substrings are
not error free. In each case we provide specific code constructions of st
rings that their reconstruction is guaranteed even in the presence of fail
ure in either model. We present efficient encoding and decoding maps and a
nalyze the cardinality of the code constructions.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture:
92990701982
For password to lecture, please contact: sagimar@cs.technion.ac.il
UID:123se24012024100200
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20210404T110000
DTEND;TZID="Asia/Jerusalem":20210404T120000
DTSTAMP;TZID="Asia/Jerusalem":20210404T110000
FREEBUSY;FBTYPE=BUSY:20210404T110000/20210404T120000
SUMMARY;LANGUAGE=en-US:phd talk by Galia Nordon about Leveraging Drug Moda
lities for Drug Repurposing at 2021-04-04 11:00:00
DESCRIPTION;LANGUAGE=en-US:Drug repurposing is the process of applying kno
wn drugs to treat new diseases. Successful repurposing can reduce costs an
d time to market as medications have already passed studies of human safet
y. It is an important task due to the length of time and the large cost of
novel drug development. In recent years, alongside the growing resources
needed for developing new drugs, large biomedical repositories are becomin
g available as well as the maturing technology for analyzing them. These f
actors make the task of drug repurposing a relevant and important one.
In this thesis we address the task of drug repurposing using three data
modalities: (1) Electronic Health Records collected for over 10 years for
over 2 million patients. (2) Biomedical literature consisting of over 28 m
illion publications. (3) The chemical structure of the drug compound.
These modalities are complementary as each of them provides a different k
ind of information. Electronic health records hold comprehensive observati
onal data, biomedical literature holds theoretical knowledge, and the chem
ical structure describes the fundamental properties of the drug.
We d
escribe the results obtained from analyzing electronic health records, reg
arding the task of drug re-purposing for Hypertension and Type II Diabetes
as well as consequent discoveries made regrading the effects of beta-bloc
kers on Parkinson's morbidity. We discuss the challenges in such an analys
is and continue to demonstrate how combining literature knowledge may aid
this task. We further build a medical-condition causal graph based on thes
e two repositories.
We then demonstrate the use of the chemical struc
ture modality: alone, for the task of lead optimization, and combined with
biomedical literature, for the task of embedding drugs in a vector space.
We show the embedding we obtain is useful in predicting drug repurposing.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture:
96920869630
For password to lecture, please contact: galiasn@cs.technion.ac.il
UID:123se24012024100180
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20210405T123000
DTEND;TZID="Asia/Jerusalem":20210405T133000
DTSTAMP;TZID="Asia/Jerusalem":20210405T123000
FREEBUSY;FBTYPE=BUSY:20210405T123000/20210405T133000
SUMMARY;LANGUAGE=en-US:Theory of Data talk by Supratim Shit (Indian Instit
ute of Technology Gandhinagar) about Data Science & Deep Learning: Coreset
s for Some Machine Learning Algorithms at 2021-04-05 12:30:00
DESCRIPTION;LANGUAGE=en-US:A butterfly network consists of logarithmically
many layers, each with a linear number of pre-specified nonzero weights.
We propose to replace a dense linear layer in any neural network by an arc
hitecture based on the butterfly network. The proposed architecture signif
icantly improves upon the quadratic number of weights required in a standa
rd dense layer to nearly linear with little compromise in expressibility o
f the resulting operator. In a collection of wide variety of experiments,
including supervised prediction on both the NLP and vision data, we show t
hat this not only produces results that match and often outperform existin
g well-known architectures, but it also offers faster training and predict
ion in deployment.
Theoretical result presented in the paper explain
why the training speed and outcome are not compromised by our proposed ap
proach.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture:
98095992835
UID:123se24012024100320
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20210405T173000
DTEND;TZID="Asia/Jerusalem":20210405T183000
DTSTAMP;TZID="Asia/Jerusalem":20210405T173000
FREEBUSY;FBTYPE=BUSY:20210405T173000/20210405T183000
SUMMARY;LANGUAGE=en-US:msc talk by Daniella Bar-Lev about The Deletion/In
sertion Channel and its Application to Coding for DNA Storage at 2021-04-0
5 17:30:00
DESCRIPTION;LANGUAGE=en-US:DNA-based storage offers significant advantages
over magnetic and optical storage solutions in terms of density, durabili
ty and not requiring a constant power supply. Given current trends of cost
reduction in DNA synthesis and sequencing, it is now acknowledged that wi
thin the next 10 – 15 years DNA-based storage may become a highly competit
ive archiving technology.
The microscopic world in which the DNA mo
lecules reside induces error patterns that are fundamentally different fro
m their digital counterparts. Errors in DNA are typically substitutions, d
uplications, insertions and deletions. These types of errors result from t
he specific error behavior in DNA and the method in which DNA strands are
stored together. Currently, in most DNA storage systems, the coding soluti
ons used for error correction were conventional ones such as repetition, R
eed-Solomon and LDPC codes. Hence, to maintain reliability in reading and
writing, new coding schemes must be developed. This work addresses some of
the open problems regarding coding for DNA storage and more generally cod
ing for channels with deletions.
When the number of deletions is equ
al to the number of insertions, the Levenshtein metric is a suitable measu
re to compute the distance between two words of the same length. We presen
t the minimum, maximum, and average size of a ball with radius one, in the
Levenshtein metric. The related minimum and maximum size of a maximal ant
icode with diameter one is also calculated.
Beyond the analysis of th
e sizes of algebraic structures relevant to the study of DNA-based storage
, the reconstruction problem is also discussed. That is, to reconstruct a
word given several of its noisy copies, which is relevant to several appli
cations. The goal in the DNA reconstruction problem is to minimize the dis
tance between the original sequence and the algorithm's estimation. We con
sider two variants of the reconstruction problem and constructed an optima
l decoder in terms of average error probability for both of them.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture:
97428473352
For password to lecture, please contact: daniellalev@cs.technion.ac.il
UID:123se24012024100250
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20210405T173000
DTEND;TZID="Asia/Jerusalem":20210405T193000
DTSTAMP;TZID="Asia/Jerusalem":20210405T173000
FREEBUSY;FBTYPE=BUSY:20210405T173000/20210405T193000
SUMMARY;LANGUAGE=en-US:CSpecial Event about Recruitment Day By NVDIA at 20
21-04-05 17:30:00
DESCRIPTION;LANGUAGE=en-US:NVDIA will hold a TEAMS online recruitment day today, Monday, Apri
l 5th, 2021, including short technological lectures and Q&A meeting wi
th the company's engineers, job offers and information about openings.
You are all invited.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:TEAMS Event
UID:123se24012024100350
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20210406T090000
DTEND;TZID="Asia/Jerusalem":20210406T100000
DTSTAMP;TZID="Asia/Jerusalem":20210406T090000
FREEBUSY;FBTYPE=BUSY:20210406T090000/20210406T100000
SUMMARY;LANGUAGE=en-US:pixel-club talk by Yi Ma (University of California,
Berkeley) about Pixel Club: Deep Networks from First Principles at 2021-0
4-06 09:00:00
DESCRIPTION;LANGUAGE=en-US:In this talk, we offer an entirely “white box’’
interpretation of deep (convolution) networks from the perspective of dat
a compression (and group invariance). In particular, we show how modern de
ep layered architectures, linear (convolution) operators and nonlinear act
ivations, and even all parameters can be derived from the principle of max
imizing rate reduction (with group invariance). All layers, operators, and
parameters of the network are explicitly constructed via forward propagat
ion, instead of learned via back propagation. All components of so-obtaine
d network, called ReduNet, have precise optimization, geometric, and stati
stical interpretation. There are also several nice surprises from this pri
ncipled approach: it reveals a fundamental tradeoff between invariance and
sparsity for class separability; it reveals a fundamental connection betw
een deep networks and Fourier transform for group invariance – the computa
tional advantage in the spectral domain (why spiking neurons?); this appro
ach also clarifies the mathematical role of forward propagation (optimizat
ion) and backward propagation (variation). In particular, the so-obtained
ReduNet is amenable to fine-tuning via both forward and backward (stochast
ic) propagation, both for optimizing the same objective.
This is join
t work with students Yaodong Yu, Ryan Chan, Haozhi Qi of Berkeley, Dr. Cho
ng You now at Google Research, and professor John Wright of Columbia Unive
rsity.
Short Bio:
Yi Ma is a Professor in residence at the Departme
nt of Electrical Engineering and Computer Sciences at the University of Ca
lifornia, Berkeley. He received his Bachelor’s degree from Tsinghua Univer
sity in 1995 and MS and PhD degrees from UC Berkeley in 2000. His research
interests are in computer vision, high-dimensional data analysis, and int
elligent systems. He has been on the faculty of UIUC ECE from 2000 to 2011
, the manager of the Visual Computing group of Microsoft Research Asia fro
m 2009to 2014, and the Dean of the School of Information Science and Techn
ology ofShanghaiTech University from 2014 to 2017. He has published over 1
60 papers and three textbooks in computer vision, statistical learning, an
d data science. He received NSF Career award in 2004 and ONR Young Investi
gator award in 2005. He also received the David Marr prize in computer vis
ion in 1999 and has served as program Chair and General Chair of ICCV 2013
and 2015, respectively. He is aFellow of IEEE, SIAM, and ACM.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture:
91767504571
UID:123se24012024100330
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20210406T110000
DTEND;TZID="Asia/Jerusalem":20210406T120000
DTSTAMP;TZID="Asia/Jerusalem":20210406T110000
FREEBUSY;FBTYPE=BUSY:20210406T110000/20210406T120000
SUMMARY;LANGUAGE=en-US:msc talk by Shaked Brody about A Structural Model f
or Contextual Code Changes at 2021-04-06 11:00:00
DESCRIPTION;LANGUAGE=en-US:We address the problem of predicting edit compl
etions based on a learned model that was trained on past edits. Given a co
de snippet that is partially edited, our goal is to predict a completion o
f the edit for the rest of the snippet. We refer to this task as the Edit
Completion task and present a novel approach for tackling it. The main ide
a is to directly represent structural edits. This allows us to model the l
ikelihood of the edit itself, rather than learning the likelihood of the e
dited code. We represent an edit operation as a path in the program's Abst
ract Syntax Tree (AST), originating from the source of the edit to the tar
get of the edit. Using this representation, we present a powerful and ligh
tweight neural model for the Edit Completion task. We conduct a thorough e
valuation, comparing our approach to a variety of representation and model
ing approaches that are driven by multiple strong models such as LSTMs, Tr
ansformers, and neural CRFs. Our experiments show that our model achieves
a 28% relative gain over state-of-the-art sequential models and 2x higher
accuracy than syntactic models that learn to generate the edited code, as
opposed to modeling the edits directly.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture:
96914709680
For password to lecture, please contact: shakedbr@cs.technion.ac.il
UID:123se24012024100170
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20210407T113000
DTEND;TZID="Asia/Jerusalem":20210407T123000
DTSTAMP;TZID="Asia/Jerusalem":20210407T113000
FREEBUSY;FBTYPE=BUSY:20210407T113000/20210407T123000
SUMMARY;LANGUAGE=en-US:msc talk by Igor Margulis about On Anomaly Detectio
n in Tabular Data at 2021-04-07 11:30:00
DESCRIPTION;LANGUAGE=en-US:Anomaly detection is a technique for finding un
usual patterns in the given data.
The study of anomaly detection has
a long history and spans multiple disciplines including engineering, machi
ne learning, statistics and real-life applications.
We consider the p
roblem of anomaly detection in tabular data, and present a detection schem
e which is based on training a multiway classification model for discrimin
ating between dozens of transformations applied to given "normal" records.
The auxiliary expertise learned by the model generates feature repre
sentations that allow, at test time, identification of anomalous records b
ased on the placement of representations of transformed test records with
respect to the learned representations.
We apply the scheme while in
corporating the recently proposed classification models designed for tabul
ar data and present the results obtained for the real-life data.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture: 91383403107
For password to lecture, please contact: margulis@campus.technion.ac.il
UID:123se24012024100300
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20210408T163000
DTEND;TZID="Asia/Jerusalem":20210408T173000
DTSTAMP;TZID="Asia/Jerusalem":20210408T163000
FREEBUSY;FBTYPE=BUSY:20210408T163000/20210408T173000
SUMMARY;LANGUAGE=en-US:msc talk by Gal Peretz about What If: Answer Simula
tion Questions by Generating Code at 2021-04-08 16:30:00
DESCRIPTION;LANGUAGE=en-US:Many texts, especially in Chemistry and Biol-og
y, describe complex processes. To answer questions about such process
es one needs to understand the interactions between the different entit
ies and to track the state transition between the different stages of the
process. In this work, we tackle this problem by learning to generate co
rresponding code to a text that describes a chemical reaction process
and a question that asks about the process outcome in a different set
up. We define a domain-specific-language for such processes, and contri
bute to the community a unique dataset, curated by chemists, of proces
s texts, simulation questions, and their corresponding codes. We propos
e a reinforcement-learning based approach to learn to generate cod
e based on texts and questions optimizing both for syntactic code
similarity and the semantic run-time similarity.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture: 98204535821
For password to lecture, please contact: sgalprz@cs.technion.ac.il
UID:123se24012024100290
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20210411T110000
DTEND;TZID="Asia/Jerusalem":20210411T120000
DTSTAMP;TZID="Asia/Jerusalem":20210411T110000
FREEBUSY;FBTYPE=BUSY:20210411T110000/20210411T120000
SUMMARY;LANGUAGE=en-US:msc talk by Dolev Elbaz about Complex Event Forecas
ting in Multivariate Time Series at 2021-04-11 11:00:00
DESCRIPTION;LANGUAGE=en-US:Time-series forecasting is widely employed in a
variety of domains to predict future trends, tendencies, and properties o
f the data. However, predicting simple data items is often not enough. Man
y applications are characterized by a requirement to simultaneously monito
r hundreds or even thousands of data series and could benefit from recogni
zing future occurrences of composite patterns in advance. Despite the risi
ng need for such functionality, this problem received limited attention in
recent years.
In this work, we formally define and study the problem
of predicting patterns over basic data items in multivariate time-series
data. Our proposed solution utilizes a combination of a deep learning time
-series forecasting model and a complex event processing (CEP) evaluation
tree. We also apply attention mechanisms to improve the performance of the
forecasting models.
We devise a system capable of forecasting compos
ite patterns in a multivariate time-series using a variety of models and s
uitable for multiple types of data. Our extensive experimental evaluation
on three real-world datasets demonstrates the effectiveness and accuracy o
f our approach.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture:
996692671429
For password to lecture, please contact: dolevelb@campus.technion.ac.il
UID:123se24012024100340
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20210411T173000
DTEND;TZID="Asia/Jerusalem":20210411T193000
DTSTAMP;TZID="Asia/Jerusalem":20210411T173000
FREEBUSY;FBTYPE=BUSY:20210411T173000/20210411T193000
SUMMARY;LANGUAGE=en-US:CSpecial Event about Code Retreat Workshop at CS at
2021-04-11 17:30:00
DESCRIPTION;LANGUAGE=en-US:You are invited to participate in the Code Retr
eat workshop that will take place at CS for the first time, in order sharp
en the development skill and practice a four-hand programming method in fo
ur hands and one keyboard (Pair Programming), during which participants pr
actice writing code in pairs and sharpen code skills while coordinating gr
oup work expectations, dealing with one thought, explore and practice diff
erent methods of software development:
• Starting with a simple pro
gramming problem.
• Working in a team with another partner to discuss an
d solve the problem
• More restrictions are added to force you to consid
er different approaches
• After each round, deleting the code and return
ing to the beginning
The event will take place on Sunday, April 11,
2021, 17:30 online zoom for about two hours, in the first 10 minutes an ex
planation of the activity will be given, then a split into rooms to start
work, and participants will work in pairs throughout the activity with gui
dance and guidance from the workshop instructor.
A link to the meetin
g will be sent after pre-registration (register in pairs!)
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Event: Registration
UID:123se24012024100310
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20210412T103000
DTEND;TZID="Asia/Jerusalem":20210412T113000
DTSTAMP;TZID="Asia/Jerusalem":20210412T103000
FREEBUSY;FBTYPE=BUSY:20210412T103000/20210412T113000
SUMMARY;LANGUAGE=en-US:cggc talk by Niloy J. Mitra (University College Lon
don (UCL)) about CGGC Seminar: Deep 3D Generative Modeling at 2021-04-12 1
0:30:00
DESCRIPTION;LANGUAGE=en-US:Deep learning has taken the Computer Graphics w
orld by storm. While remarkable progress has been reported in the context
of supervised learning, the state of unsupervised learning, in contrast, r
emains quite primitive. In this talk, we will discuss recent advances wher
e we have combined knowledge from traditional computer graphics and image
formation models to enable deep generative modeling workflows. We will des
cribe how we have combined modeling and rendering, in the unsupervised set
ting, to enable controllable and realistic image and animation production.
The work is done in collaboration with various students and research coll
eagues.
Interested parties can email gershon@cs.technion.ac.il or mirela@cs.technion.ac.il for the zoom link.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture: 91344952941
For password to lecture please contact inbalb@cs.technion.ac.il
UID:123se24012024100280
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20210412T123000
DTEND;TZID="Asia/Jerusalem":20210412T133000
DTSTAMP;TZID="Asia/Jerusalem":20210412T123000
FREEBUSY;FBTYPE=BUSY:20210412T123000/20210412T133000
SUMMARY;LANGUAGE=en-US:msc talk by Omer Leibovitch about Sparse Linear Net
works with a Fixed Butterfly at 2021-04-12 12:30:00
DESCRIPTION;LANGUAGE=en-US:A butterfly network consists of logarithmically
many layers, each with a linear number of pre-specified nonzero weights.
We propose to replace a dense linear layer in any neural network by an arc
hitecture based on the butterfly network. The proposed architecture signif
icantly improves upon the quadratic number of weights required in a standa
rd dense layer to nearly linear with little compromise in expressibility o
f the resulting operator. In a collection of wide variety of experiments,
including supervised prediction on both the NLP and vision data, we show t
hat this not only produces results that match and often outperform existin
g well-known architectures, but it also offers faster training and predict
ion in deployment.
Theoretical result presented in the paper explain
why the training speed and outcome are not compromised by our proposed ap
proach.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture:
98712430421
For password to lecture, please contact: mayasidis@campus.technion.ac.il
UID:123se24012024100370
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20210412T123000
DTEND;TZID="Asia/Jerusalem":20210412T133000
DTSTAMP;TZID="Asia/Jerusalem":20210412T123000
FREEBUSY;FBTYPE=BUSY:20210412T123000/20210412T133000
SUMMARY;LANGUAGE=en-US:Theory of Data talk by Omer Leibovitch (CS, Technio
n) about Data Science & Deep Learning: Sparse Linear Networks with a Fixed
Butterfly at 2021-04-12 12:30:00
DESCRIPTION;LANGUAGE=en-US:A butterfly network consists of logarithmically
many layers, each with a linear number of pre-specified nonzero weights.
We propose to replace a dense linear layer in any neural network by an arc
hitecture based on the butterfly network. The proposed architecture signif
icantly improves upon the quadratic number of weights required in a standa
rd dense layer to nearly linear with little compromise in expressibility o
f the resulting operator. In a collection of wide variety of experiments,
including supervised prediction on both the NLP and vision data, we show t
hat this not only produces results that match and often outperform existin
g well-known architectures, but it also offers faster training and predict
ion in deployment.
Theoretical result presented in the paper explain
why the training speed and outcome are not compromised by our proposed ap
proach.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture:
98712430421
For password to lecture, please contact: mayasidis@cs.technion.ac.il
UID:123se24012024100380
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20210418T090000
DTEND;TZID="Asia/Jerusalem":20210418T100000
DTSTAMP;TZID="Asia/Jerusalem":20210418T090000
FREEBUSY;FBTYPE=BUSY:20210418T090000/20210418T100000
SUMMARY;LANGUAGE=en-US:phd talk by Tomer Golany about Deep Generative Mode
ls for ECG Classification at 2021-04-18 09:00:00
DESCRIPTION;LANGUAGE=en-US:32% of all global deaths in the world are cause
d by cardiovascular diseases. The Electrocardiogram (ECG) is a non-invasiv
e tool to measure the electrical activity of the heart, and it is the most
common test performed by cardiologists to detect heart-diseases.
Ana
lyzing ECG signals manually is a hard task. Furthermore, abnormalities in
the heart may occur at any time and not necessarily in the hospital.
Ma
ny attempts were made to automate this task using machine learning algorit
hms. However, this task is challenging. There are many types of possible d
isorders in the heart and extreme variations exist between different patie
nts. Furthermore, for a deep learning model to succeed in its task, a larg
e amount of annotated data is needed. However, analyzing and labeling ECG
signals manually is prone to errors and consumes expensive experts time. W
e study the use of generative adversarial networks (GANs) for the synthesi
s of ECG signals, which can then be used as additional training data to im
prove ECG classification.
First, we present a GAN that learns to synt
hesize ECG heartbeats which can then be used as additional training data t
o improve classifier performance. Furthermore, to synthesize patient-speci
fic ECG heartbeats, we propose a modified GAN optimized using a specialize
d loss function to mimic the morphology of the subject’s cardiac signal.
In the second part of the thesis we take advantage of the nature of
the ECG signal which describes the biological system of the heart. We stu
dy how to incorporate this knowledge into the generative process by levera
ging a biological ECG simulator of the heart, defined by a system of ordin
ary differential equations (ODE) representing heart dynamics.
Furthermor
e, We study how the ECG dynamics can be learned directly by a GAN that com
bines both physical and data considerations.
We introduce an ECG-ODE
-GAN framework, in which the generator learns the dynamics of a physical s
ystem in the form of an ODE.
Finally, we study the dynamics of a full
12-lead ECG signal, the most commonly used ECG exam in medical facilities
. ECG sensors were developed to allow for the recording of the full 12-lea
d ECG signal at home. However, if even a single lead is not correctly posi
tioned on the body that lead becomes corrupted, making an automatic diagno
sis on the basis of the full signal impossible. We present a methodology t
o reconstruct missing or noisy leads using the theory of Koopman Operators
. We learn a dynamical system describing the evolution of the 12 individua
l signals together in time, which enables to impute missing leads by solvi
ng a novel least-squares system. We perform an empirical evaluation using
12-lead ECG signals from thousands of patients and show that we are able t
o reconstruct signals in such a way that enables accurate clinical diagnos
is.
This thesis is one of the first works to demonstrate the ability
of deep learning to leverage knowledge about physiological systems to gene
rate synthetic examples, specifically ECG signals. The algorithms presente
d enable full automation of ECG analysis, both in hospitals, and in home s
ensors.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture:
996761764160
For password to lecture, please contact: tomer.golany@cs.technion.ac.il
UID:123se24012024100420
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20210419T123000
DTEND;TZID="Asia/Jerusalem":20210419T133000
DTSTAMP;TZID="Asia/Jerusalem":20210419T123000
FREEBUSY;FBTYPE=BUSY:20210419T123000/20210419T133000
SUMMARY;LANGUAGE=en-US:Theory of Data talk by Vineet Nair (CS, Technion) a
bout Data Science & Deep Learning: State Visitation Fairness in Average-Re
ward MDPs at 2021-04-19 12:30:00
DESCRIPTION;LANGUAGE=en-US:Fairness has emerged as an important concern in
automated decision-making in recent years, especially when these decision
s affect human welfare. In this work, we study fairness in temporally exte
nded decision-making settings, specifically those formulated as Markov Dec
ision Processes (MDPs). Our proposed notion of fairness ensures that each
state's long-term visitation frequency is more than a specified fraction.
In an average-reward MDP setting, we formulate the problem as a bilinear s
addle point program and, for a generative model, solve it using a Stochas
tic Mirror Descent (SMD) based algorithm. The proposed solution guarantees
a simultaneous approximation of the expected average-reward and the long-
term state-visitation frequency.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture:
93378688224
For password to lecture, please contact: mayasidis@cs.technion.ac.il
UID:123se24012024100430
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20210419T170000
DTEND;TZID="Asia/Jerusalem":20210419T180000
DTSTAMP;TZID="Asia/Jerusalem":20210419T170000
FREEBUSY;FBTYPE=BUSY:20210419T170000/20210419T180000
SUMMARY;LANGUAGE=en-US:cggc talk by Yusu Wang (University of California)
about CGGC Seminar: Topological and Geometric Analysis of Graphs at 2021-0
4-19 17:00:00
DESCRIPTION;LANGUAGE=en-US:In recent years, topological and geometric data
analysis (TGDA) has emerged as a new and promising field for processing,
analyzing and understanding complex data. Indeed, geometry and topology fo
rm natural platforms for data analysis, with geometry describing the ”shap
e” behind data; and topology characterizing / summarizing both the domain
where data are sampled from, as well as functions and maps associated to t
hem.
In this talk, I will show how topological (and geometric ideas)
can be used to analyze graph data, which occurs ubiquitously across scienc
e and engineering. Graphs could be geometric in nature, such as road netwo
rks in GIS, or relational and abstract. I will particularly focus on the r
econstruction of hidden geometric graphs from noisy data, as well as graph
matching and classification. I will discuss the motivating applications,
algorithm development, and theoretical guarantees for these methods. Throu
gh these topics, I aim to illustrate the important role that topological a
nd geometric ideas can play in data analysis.
Interested parties ca e
mail gershon@cs.technion.ac.il<
/a> or mirela@cs.tech
nion.ac.il for the zoom link.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture: 91344952941
For password to lecture please contact inbalb@cs.technion.ac.il
UID:123se24012024100440
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20210419T180000
DTEND;TZID="Asia/Jerusalem":20210419T190000
DTSTAMP;TZID="Asia/Jerusalem":20210419T180000
FREEBUSY;FBTYPE=BUSY:20210419T180000/20210419T190000
SUMMARY;LANGUAGE=en-US:CSpecial Event about Recruitment Day By CISCO at 20
21-04-19 18:00:00
DESCRIPTION;LANGUAGE=en-US:CISCO will hold an online recruitment on Monday
, April 19th, 2021, 18:00, including meetings with the company's students
and engineers who will tell you about working at Cisco's various locations
and about openings.
More details, and registration.
You are all invited!
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Event: Registration
UID:123se24012024100450
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20210421T123000
DTEND;TZID="Asia/Jerusalem":20210421T143000
DTSTAMP;TZID="Asia/Jerusalem":20210421T123000
FREEBUSY;FBTYPE=BUSY:20210421T123000/20210421T143000
SUMMARY;LANGUAGE=en-US:CSpecial Event about CS Open Day for Graduate Studi
es at 2021-04-21 12:30:00
DESCRIPTION;LANGUAGE=en-US:Technion CS open day 2021 invites outstanding u
ndergraduates from all universities to learn about the Computer Science De
partment and register for Winter Semester 2021-22.
The event will be
held online by ZOOM - ID MEETING NO. 96244586510, on Wednesday, April 2
1, 2021. between 12:30-13:45.
The program will include review on curr
iculum, research and life at the Technion CS Department:
12:30-12:40 CS
Dean, Prof. Dan Geiger
12:40-12:55 Vice Dean, Prof. Gill Barequet
12:5
5-13:20 Dr. Kira Radinsky, Head of Diagnostic Robotics: Digital Healthcare
- The Next Frontier
13:20-13:30 Mr. Gil Ben-Shachar and Ms. Stav Perle
(Ph.D. students): Life in the Computer Science Faculty
13:30 Quest
ions and answers
Attendance at the open day requires pre-r
egistration.
More details and program
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Event: Registration
UID:123se24012024100080
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20210421T140000
DTEND;TZID="Asia/Jerusalem":20210421T150000
DTSTAMP;TZID="Asia/Jerusalem":20210421T140000
FREEBUSY;FBTYPE=BUSY:20210421T140000/20210421T150000
SUMMARY;LANGUAGE=en-US:msc talk by Liron Bronfman about PCPs and Cryptogra
phy: New Limitations and Opportunities at 2021-04-21 14:00:00
DESCRIPTION;LANGUAGE=en-US:The connection between information theoretic pr
oof systems and cryptography has been extremely fruitful. In this thesis,
we further explore this connection, showing both new limitations and oppor
tunities.
In the talk we will focus on the new opportunities and show
constructions of computational relaxations of objects that are known to b
e essentially impossible to achieve information theoretically. In particul
ar, we show cryptographic analogs of:
(1) PCPs whose length is proportio
nal to the witness size.
(2) Instance compression, which allows, for exa
mple, to efficiently and generically reduce the size of a given formula on
m clauses and n variables (with m >>n) to a formula of size poly(n,log(m)
).
We will discuss the applicability of these relaxations and raise q
uestions for future research.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture:
5480679598
For password to lecture, please contact: br@cs.technion.ac.il
UID:123se24012024100390
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20210421T173000
DTEND;TZID="Asia/Jerusalem":20210421T193000
DTSTAMP;TZID="Asia/Jerusalem":20210421T173000
FREEBUSY;FBTYPE=BUSY:20210421T173000/20210421T193000
SUMMARY;LANGUAGE=en-US:CSpecial Event about git Workshop at CS at 2021-04-
21 17:30:00
DESCRIPTION;LANGUAGE=en-US:You are invited to an online technological work
shop by Aviv Rosenberg,CS Ph.D. student and TA, on versioning with git: Ho
w to stop being afraid of changing code, on Wednesday, April 21, 2021, 17
:30.
More details
on the the workshop agenda and pre-registration.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Event: Registration
UID:123se24012024100460
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20210427T113000
DTEND;TZID="Asia/Jerusalem":20210427T123000
DTSTAMP;TZID="Asia/Jerusalem":20210427T113000
FREEBUSY;FBTYPE=BUSY:20210427T113000/20210427T123000
SUMMARY;LANGUAGE=en-US:pixel-club talk by Ronen Basri (Weizmann Institute
of Science) about Pixel Club: On the Connection between Deep Neural Networ
ks and Kernel Methods at 2021-04-27 11:30:00
DESCRIPTION;LANGUAGE=en-US:Recent theoretical work has shown that massivel
y overparameterized neural networks are equivalent to kernel regressors th
at use Neural Tangent Kernels (NTKs). Experiments indicate that these kern
el methods perform similarly to real neural networks. My work in this subj
ect aims to better understand the properties of NTK and relate them to pro
perties of real neural networks. In particular, I will argue that for inpu
t data distributed uniformly on the sphere NTK favors low frequency predic
tions over high frequency ones, potentially explaining why overparameteriz
ed networks do not overfit their training data. I will further discuss the
behavior of NTK when data is distributed nonuniformly, and finally show t
hat NTK is tightly related to the classical Laplace kernel, which has a si
mple closed form. Our results suggest that much insight about neural netwo
rks can be obtained from analysis of NTK.
Bio:
Ronen Basri is Pro
fessor of Computer Science and Dean of Mathematics and Computer Science at
the Weizmann Institute of Science. Additionally, he is the incumbent of t
he Elaine and Bram Goldsmith Chair of Applied Mathematics. Ronen Basri rec
eived his Ph.D. degree from the Weizmann Institute of Science, and was pos
tdoctoral fellow at the Massachusetts Institute of Technology before assum
ing a faculty position at Weizmann. He further held visiting positions at
NEC Research Institute, Toyota Technological Institute at Chicago, Howard
Hughes Janelia Farm Campus and the University of Maryland at College Park.
His research interests include computer vision and machine learning. His
recent work focuses primarily on shape modeling and 3D reconstruction, as
well as theoretical aspects of deep learning.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture:
91488539030
UID:123se24012024100490
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20210427T123000
DTEND;TZID="Asia/Jerusalem":20210427T133000
DTSTAMP;TZID="Asia/Jerusalem":20210427T123000
FREEBUSY;FBTYPE=BUSY:20210427T123000/20210427T133000
SUMMARY;LANGUAGE=en-US:colloq talk by Roee Shraga - Guest Lecture about Re
shaping the Roles of Humans and Al in Data Integration at 2021-04-27 12:3
0:00
DESCRIPTION;LANGUAGE=en-US:The matching task is at the heart of data integ
ration, in charge of aligning elements of data sources. Matching is a hand
y tool in multiple contemporary business and commerce applications and has
been investigated in the fields of databases, AI, Semantic Web, and data
mining for many years. The core challenge still remains the ability to cre
ate quality algorithmic matchers, automatic tools for identifying correspo
ndences among data concepts (e.g., database attributes). Matching problems
were traditionally performed in a semi-automatic manner, with corresponde
nces being generated by matching algorithms and outcomes subsequently vali
dated by human experts. In this talk, I will discuss the merits of human-i
n-the-loop data integration with an emphasis on the obstacles of achieving
effective human matching and validation. To illustrate the ability of mac
hine learning to support human-in-the-loop, I will present a novel charact
erization of "human matching experts" and provide a novel framework to ide
ntify reliable and valuable human experts. The framework is accompanied by
a novel set of features and was shown to be useful using an extensive emp
irical evaluation. In particular, we show that our approach can improve ma
tching results by electing expert matchers. To conclude the talk, I will e
laborate on our recent research including a deep learning mechanism to cal
ibrate and filter human matching decisions to improve the quality of a mat
ch.
BIO:
Roee Shraga is a Postdoctoral fellow at the Technion - Is
rael Institute of Technology, from which he received a PhD degree in 2020
in the area of Data Science. Roee has published more than a dozen papers i
n leading journals and conferences on the topics of data integration, huma
n-in-the-loop, machine learning, process mining, and information retrieval
. He is also a recipient of several PhD fellowships including the Leonard
and Diane Sherman Interdisciplinary Fellowship (2017), the Daniel Excellen
ce Scholarship (2019), and the Miriam and Aaron Gutwirth Memorial Fellowsh
ip (2020).
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:HYBRID - Taub 5 (Green Pass) and
Zoom Lecture:
91488539030
UID:123se24012024100510
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20210429T143000
DTEND;TZID="Asia/Jerusalem":20210429T153000
DTSTAMP;TZID="Asia/Jerusalem":20210429T143000
FREEBUSY;FBTYPE=BUSY:20210429T143000/20210429T153000
SUMMARY;LANGUAGE=en-US:msc talk by Lior Ben-Yamin about Maximizing Through
put in Flow Shop Real-time Scheduling at 2021-04-29 14:30:00
DESCRIPTION;LANGUAGE=en-US:We consider scheduling real-time jobs in the cl
assic flow shop model. The input is a set of n jobs, each consisting of m
segments to be processed on m machines in the specified order. Each job al
so has a release time, a due date, and a weight. The objective is to maxim
ize the throughput, i.e., to find a subset of the jobs that have the maxim
um total weight and can complete processing on the m machines within their
time windows. This problem has numerous real-life applications ranging fr
om manufacturing to cloud and embedded computing platforms, already in the
special case where m=2.
Previous work in the flow shop model has foc
used on makespan, flow time, or tardiness objectives. However, little is k
nown for the flow shop model in the real-time setting. In this work, we gi
ve the first nontrivial results for this problem and present a pseudo-poly
nomial time (2m+1)-approximation algorithm for throughput maximization on
$m \geq 2$ machines, where m is a constant. This ratio is essentially tigh
t due to a known hardness of approximation result. For the two-machine cas
e, we give a polynomial-time $9+\eps$-approximation algorithms, where $\ep
s = O(1/n)$. Better bounds are derived for some restricted subclasses of i
nputs with two machines, as well as the no-wait flow shop model.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture:
93508538152
For password to lecture, please contact: lior.b@cs.technion.ac.il
UID:123se24012024100410
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20210429T153000
DTEND;TZID="Asia/Jerusalem":20210429T163000
DTSTAMP;TZID="Asia/Jerusalem":20210429T153000
FREEBUSY;FBTYPE=BUSY:20210429T153000/20210429T163000
SUMMARY;LANGUAGE=en-US:phd talk by Shoval Lagziel about Computational infe
rence of cancer metabolic alterations for early diagnosis and treatment at
2021-04-29 15:30:00
DESCRIPTION;LANGUAGE=en-US:Metabolic reprogramming is a hallmark of cancer
, providing novel means to selectively target cancer cells, for precision
medicine and early diagnosis. Understanding tumor-specific metabolic alter
ations facilitates the identification of induced dependency on specific en
zymes whose inhibition selectively targets cancer cells. In addition, the
altered metabolic activity of cancer cells, involving the consumption of m
etabolic nutrients and the secretion of byproducts from the tumor leaves m
etabolic traces that can be utilized for diagnostic purposes. Here, we ex
plored two main directions based on the metabolic reprogramming of cancer:
(1) construction of models suggesting potential metabolic mechanisms for
the dependency on metabolic genes, (2) early cancer diagnosis based on fas
t and sensitive metabolomics of blood samples.
Genome-wide RNAi and C
RISPR screens are powerful tools for identifying genes essential for cance
r proliferation and survival. Previous works integrated loss-of-function s
creens with cancer cell line molecular characterization to reveal the unde
rlying mechanisms for cancer dependence on specific genes; however, explai
ning cancer dependence on metabolic genes was shown to be especially chall
enging. Considering that metabolic activity is highly dependent on nutrien
t availability, analyzing publicly available omics datasets, we have shown
that utilizing different media types for culturing cancer cell lines has
a major effect on intracellular metabolite levels and metabolic gene depen
dencies – calling for future analyses of published omics datasets such a
s that of the CCLE to account for this confounding effect. Considering cul
ture media as well as accounting for molecular features of functionally re
lated metabolic enzymes in a metabolic network enabled us to improve our u
nderstanding of cancer cell line-specific dependence on metabolic genes us
ing machine learning models.
Early diagnosis of cancer greatly increa
ses the chances for successful treatment of cancer. Major ongoing efforts
are made to develop highly sensitive, cost-effective screening methods via
a variety of molecular biomarkers. Mass spectrometry based metabolomics i
s a widely used approach in biomedical research. However, current methods
coupling mass spectrometry with chromatography are time-consuming and not
suitable for high-throughput analysis of thousands of samples. An alternat
ive approach is flow-injection mass spectrometry (FI-MS) in which samples
are directly injected into the ionization source. However, it was previous
ly shown to provide a reduced sensitivity and reproducibility. We develope
d two rapid mass-spectrometry based metabolomics methods, FI-MS based and
LC-MS based, enabling a reproducible detection and quantitation of thousan
ds of metabolites within less than one minute per sample. The developed ap
proach facilitates high-throughput metabolomics for a variety of applicati
ons, including biomarker discovery and functional genomics screens. Applyi
ng the developed metabolomics method to hundreds of serum samples from can
cer patients and healthy controls, utilizing machine learning techniques,
we have demonstrated the potential and applicability of this approach for
population-wide cancer screening.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture:
93506187830
For password to lecture, please contact: shovall@cs.technion.ac.il
UID:123se24012024100480
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20210429T170000
DTEND;TZID="Asia/Jerusalem":20210429T190000
DTSTAMP;TZID="Asia/Jerusalem":20210429T170000
FREEBUSY;FBTYPE=BUSY:20210429T170000/20210429T190000
SUMMARY;LANGUAGE=en-US:CSpecial Event about Lecture on Quantum Calculation
: What is it and why is it Cool? at 2021-04-29 17:00:00
DESCRIPTION;LANGUAGE=en-US:You are invited to a lecture on quantum computi
ng: what is it and why is it cool? by Dr. Gadi Alexandrovich - CS graduate
, a researcher in the IBM research laboratory in Haifa in the field of qua
ntum computing and the author of the "Inaccurate" mathematical b
log - which will deal with quantum computers and the changes they will bri
ng about in the future.
The lecture will take place on Thursday, Apri
l 29, 17:00, in a zoom session - a link will be sent after pre-registration.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Event: Registration
UID:123se24012024100520
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20210503T113000
DTEND;TZID="Asia/Jerusalem":20210503T123000
DTSTAMP;TZID="Asia/Jerusalem":20210503T113000
FREEBUSY;FBTYPE=BUSY:20210503T113000/20210503T123000
SUMMARY;LANGUAGE=en-US:cggc talk by Fady Massarwi (CS, Technion) about Geo
metrical Challenges in Treating Irregular Heart Beat at 2021-05-03 11:30:0
0
DESCRIPTION;LANGUAGE=en-US:This talk presents some of the geometrical aspe
cts involved in treating irregular heart beat rhythm (Arrythmia) using Car
to 3 System. Carto 3 is a product of Biosense-Webster, a global leader in
the science of diagnosing and treating heart rhythm disorders. CARTO 3 Sys
tem enables accurate visualization of multiple catheters in a patient’s he
art and pinpoints exact location/orientation of a catheter. During arrythm
ia procedure, a 3D electro-anatomical reconstruction of the heart is built
and color coded with the electrical activity in the heart. In this talk,
we’ll introduce mesh processing algorithms and discuss industrial challeng
es encountered in the process of building and coloring geometrical reconst
ructions of the heart.
Interested parties ca email gershon@cs.technion.ac.il or mirela@cs.technion.ac.il for the
zoom link.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture: 91344952941
For password to lecture please contact inbalb@cs.technion.ac.il
UID:123se24012024100470
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20210503T170000
DTEND;TZID="Asia/Jerusalem":20210503T190000
DTSTAMP;TZID="Asia/Jerusalem":20210503T170000
FREEBUSY;FBTYPE=BUSY:20210503T170000/20210503T190000
SUMMARY;LANGUAGE=en-US:CSpecial Event about Intel Ergonomics Workshop to U
pgrade the Study Position at 2021-05-03 17:00:00
DESCRIPTION;LANGUAGE=en-US:Intel representative, an expert in ergonomics,
will hold a workshop on the subject on Monday, May 3, 2021, at 17:00, with
tips for upgrading the distance learning environment.
A link to Part
icipants will be sent after pre-registration
a>.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Event: Registration
UID:123se24012024100530
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20210505T113000
DTEND;TZID="Asia/Jerusalem":20210505T123000
DTSTAMP;TZID="Asia/Jerusalem":20210505T113000
FREEBUSY;FBTYPE=BUSY:20210505T113000/20210505T123000
SUMMARY;LANGUAGE=en-US:phd talk by Rotem Liss about Security of Quantum K
ey Distribution Protocols at 2021-05-05 11:30:00
DESCRIPTION;LANGUAGE=en-US:The counter-intuitive features of quantum mecha
nics make it possible to solve problems and perform tasks that are beyond
the abilities of non-quantum (classical) computers and communication devic
es. In particular, quantum key distribution (QKD) protocols allow two part
icipants (Alice and Bob) to achieve the classically-impossible task of gen
erating a secret shared key even if their adversary is computationally unl
imited.
Unfortunately, the security promises of QKD are true only in the
ory; practical implementations of QKD deviate from the theoretical protoco
ls, and many of these deviations give rise to practical attacks.
In t
his talk, we study the security properties of various QKD protocols in man
y practical settings:
- First, we discuss practical attacks, and we show
how an important practical attack (named "Bright Illumination") can be mo
deled as a theoretical "Reversed-Space" attack.
- Then, we discuss pract
ical security of semiquantum key distribution (SQKD) protocols, where eith
er Alice or Bob is non-quantum (classical). We suggest a new SQKD protocol
(the "Mirror protocol") which can be securely implemented, and we prove i
t robust and secure against a wide range of attacks (the "collective attac
ks").
- Finally, we study "composable security" of the first QKD protoco
l created by Bennett and Brassard (BB84). BB84 has its unconditional secur
ity proved against adversaries performing the most general attacks in a th
eoretical (idealized) setting. We generalize an algebraic security approac
h for BB84 to make it prove "composable security": namely, prove that the
secret key remains secret even when Alice and Bob actually use it for cryp
tographic purposes.
Overall, the research presented in this talk aims
to enhance our understanding on how to bridge the gap between theory and
practice in various sub-fields of QKD, and it may help construct realistic
QKD implementations that can be proved truly and unconditionally secure.
The talk is self-contained and requires no prior knowledge of quantum
information.
The talk will be given in English.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture:
99607663751
For password to lecture, please contact: rotemliss@cs.technion.ac.il
UID:123se24012024100500
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20210505T113000
DTEND;TZID="Asia/Jerusalem":20210505T123000
DTSTAMP;TZID="Asia/Jerusalem":20210505T113000
FREEBUSY;FBTYPE=BUSY:20210505T113000/20210505T123000
SUMMARY;LANGUAGE=en-US:ceClub talk by Noam Bloch (VP HW architecture, NVID
IA) about ceClub: The Technion Computer Engineering Club at 2021-05-05 11:
30:00
DESCRIPTION;LANGUAGE=en-US:In Modern data centers, resources are usually v
irtualized. Applications running on those date centers are distributed ove
r many virtual machines. For those applications, the data centers provide
software defined infrastructure services for networking, storage, and secu
rity. When software defined services are running within the same CPU as th
e applications, they consume CPU resources on the expanse of the applicati
ons. Moreover, the data center security can be jeopardized
NVIDIA Da
ta Center Infrastructure Processing Unit (DPU) allow migrating all the inf
rastructure services from the application CPU. This new architecture isola
tes the data center infrastructure from the application domain running on
the CPUs or GPUs making this sensitive code independent on any application
. NVIDIA DPU accelerates and abstract the storage, networking, and securit
y service to enable applications to get the most out of the CPUs and GPUs.
In this presentation, the fundamentals of NVIDIA DPU architecture wi
ll be presented, along with the NVIDIA DOCA SDK programming mode for the D
PU, the system use cases and the impacts of this architecture on next gene
ration data centers.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture:
93093927833
UID:123se24012024100540
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20210512T113000
DTEND;TZID="Asia/Jerusalem":20210512T123000
DTSTAMP;TZID="Asia/Jerusalem":20210512T113000
FREEBUSY;FBTYPE=BUSY:20210512T113000/20210512T123000
SUMMARY;LANGUAGE=en-US:ceClub talk by Ameer Haj Ali (UC Berkeley) about ce
Club: Machine Learning in Compiler Optimization at 2021-05-12 11:30:00
DESCRIPTION;LANGUAGE=en-US:The end of Moore's law is driving the search fo
r new techniques to improve system performance as applications continue to
evolve rapidly and computing power demands continue to rise. One promisin
g technique is to build more intelligent compilers.
Compilers map high-l
evel programs to lower-level primitives that run on hardware. During this
process, compilers perform many complex optimizations to boost the perform
ance of the generated code. These optimizations often require solving NP-H
ard problems and dealing with an enormous search space. To overcome these
challenges, compilers currently use hand-engineered heuristics that can ac
hieve good but often far-from-optimal performance. Alternatively, software
engineers resort to manually writing the optimizations for every section
in the code, a burdensome process that requires prior experience and signi
ficantly increases the development time.
In this work, novel approach
es for automatically handling complex compiler optimization tasks are expl
ored. End-to-end solutions using deep reinforcement learning and other mac
hine learning algorithms are proposed. These solutions dramatically reduce
the search time while capturing the code structure, different instruction
s, dependencies, and data structures to enable learning a sophisticated mo
del that can better predict the actual performance cost and determine supe
rior compiler optimizations. The proposed techniques can outperform existi
ng state-of-the-art solutions while requiring shorter search time. Further
more, unlike existing solutions, the deep reinforcement learning solutions
are shown to generalize well to real benchmarks.
Bio:
Ameer Haj-Al
i completed his Ph.D. in Electrical Engineering and Computer Science at UC
Berkeley in two years, where he was advised by Professors Ion Stoica (RIS
E Lab) and Krste Asanovic (ADEPT Lab). At UC Berkeley Ameer helped bring u
p/led many projects spanning machine learning in compiler optimization and
hardware-software codesign. This includes Gemmini, AutoPhase, NeuroVector
izer, ProTuner, Ansor, AutoCkt, and RLDRM (awarded best paper award). Befo
re attending UC Berkeley, Ameer finished his M.Sc. studies (summa cum laud
e, the valedictorian) at the Technion in 2018, where he worked with Profes
sor Shahar Kvatinsky on using emerging memory technologies to enhance the
performance of modern computer systems.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture:
92254734234
UID:123se24012024100560
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20210513T160000
DTEND;TZID="Asia/Jerusalem":20210513T170000
DTSTAMP;TZID="Asia/Jerusalem":20210513T160000
FREEBUSY;FBTYPE=BUSY:20210513T160000/20210513T170000
SUMMARY;LANGUAGE=en-US:msc talk by Gal Benmocha about Unintended Features
of APIs: Cryptanalysis of Incremental HMAC at 2021-05-13 16:00:00
DESCRIPTION;LANGUAGE=en-US:Many cryptographic APIs provide extra functiona
lity that was not intended by the designers. In this seminar we discuss su
ch an unintended functionality in the API of HMAC as implemented by Siemen
s and OpenSSL.
HMAC authenticates a single message at a time with a
single authentication tag. However, most HMAC implementations do not compl
ain when extra data is added to the stream after a tag is computed. We cal
l such primitives Incremental MACs.
Though HMAC is not intended to b
e called incrementally, it appears that some applications (e.g., Siemens S
7 protocol) use the standard HMAC API to provide an incremental MAC. We ob
serve that calling most standard HMAC implementations incrementally did no
t take into consideration that they might be called incrementally, and thu
s cause unfortunate side-effects during tag computation.
We show that
due to these side-effects, the Siemens and OpenSSL implementations are no
t as secure as HMAC.
We also discuss other results from my research.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture: 99410484579
UID:123se24012024100550
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20210530T110000
DTEND;TZID="Asia/Jerusalem":20210530T120000
DTSTAMP;TZID="Asia/Jerusalem":20210530T110000
FREEBUSY;FBTYPE=BUSY:20210530T110000/20210530T120000
SUMMARY;LANGUAGE=en-US:msc talk by Ido Imanuel about Neural Algorithms for
Precise Shape Completion at 2021-05-30 11:00:00
DESCRIPTION;LANGUAGE=en-US:According to Aristotle, “the whole is greater t
han the sum of its parts”. This statement was adopted to explain human per
ception by the Gestalt psychology school of thought in the twentieth centu
ry. Here, we claim that when observing a part of an object which was previ
ously acquired as a whole, one could deal with both partial correspondence
and shape completion in a holistic manner. More specifically, given the ge
ometry of a full, articulated object in a given pose, as well as a partial
scan of the same object in a different pose, we address the new problem of
matching the part to the whole while simultaneously reconstructing the ne
w pose from its partial observation. Our approach is data-driven and takes
the form of a Siamese autoencoder without the requirement of a consistent
vertex labeling at inference time; as such, it can be used on unorganized
point clouds as well as on triangle meshes. We demonstrate the practical
effectiveness of our model in the applications of single-view deformable sh
ape completion and dense shape correspondence, both on synthetic and real-
world geometric data, where we outperform prior work by a large margin.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture: 8335241961
UID:123se24012024100600
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20210531T180000
DTEND;TZID="Asia/Jerusalem":20210531T200000
DTSTAMP;TZID="Asia/Jerusalem":20210531T180000
FREEBUSY;FBTYPE=BUSY:20210531T180000/20210531T200000
SUMMARY;LANGUAGE=en-US:CSpecial Event about Online Meeting with Intel on J
ob Interviews at 2021-05-31 18:00:00
DESCRIPTION;LANGUAGE=en-US:You are invited to an online meeting with Intel
representatives and to hear from their software engineers on the work exp
erience as interviewers, including tools, exercising and tips for success
in the technical stage of an interview, on Monday, May 31, 2021, 18:00.
Link to the meeting will be sent upon pre-r
egistration.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Event: Registration
UID:123se24012024100630
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20210601T103000
DTEND;TZID="Asia/Jerusalem":20210601T113000
DTSTAMP;TZID="Asia/Jerusalem":20210601T103000
FREEBUSY;FBTYPE=BUSY:20210601T103000/20210601T113000
SUMMARY;LANGUAGE=en-US:msc talk by Michael Leybovich about ML Based Lineag
e in Databases at 2021-06-01 10:30:00
DESCRIPTION;LANGUAGE=en-US:There has been extensive research on data prove
nance. Previous works were concerned with annotating the results of databa
se (DB) queries with provenance information which is useful in explaining
query results at various resolution levels. In this work, we track the lin
eage of tuples throughout their database lifetime. That is, we consider a
scenario in which tuples (records) that are produced by a query may affect
other tuple insertions into the DB, as part of a normal workflow. As time
goes on, provenance explanations for such tuples become deeply nested, in
creasingly consuming space, and resulting in decreased clarity and readab
ility.
We present a novel approach for approximating lineage tracking.
We use Machine Learning (ML) and Natural Language Processing (NLP) techniq
ues; mainly, word embedding.
The basic idea is summarizing (and approxim
ating) the lineage of each tuple via a small set of constant-size vectors
(the number of vectors per-tuple is a hyperparameter). For explicitly (and
independently of DB contents) inserted tuples - the vectors are obtaine
d via a pre-trained word vectors model over their underlying database doma
in “text”. During the execution of a query, we construct the lineage vecto
rs of the final (and intermediate) result tuples in a similar fashion to t
hat of semiring-based exact provenance calculations. We extend the + and *
operations to generate sets of lineage vectors, while emphasizing the abi
lity to propagate information and preserve the compact representation. The
refore, our solution does not suffer from space complexity blow-up over ti
me. Another significant benefit of our approach is a "natural ranking" of
explanations to the existence of a tuple in the DB.
We devise a genetic
s-inspired improvement to our basic method. The columns of an entity are a
tuple’s basic properties, i.e., the “genes” that combine to form its gene
tic code. In this setting, finding the lineage of a tuple in the DB is ana
logous to finding its predecessors via DNA examination. We design an alter
native lineage tracking mechanism, that of keeping track of and querying l
ineage (via embeddings) at the column (“gene”) level. This way, we manage
to better distinguish between the provenance features and the textual char
acteristics of a tuple.
We further introduce several improvements and e
xtensions to the basic method:
* Emphasizing important columns with a que
ry-dependent column weighting.
* Filtering non-contributing tuples using
Bloom Filters of queries.
* Filtering non-contributing tuples by a tuple
creation timestamp.
* Similarity weighting with query dependency DAG.
*
Extending the method to track where provenance.
We integrate our lineag
e computations into the PostgreSQL system via an extension (ProvSQL) and e
xperimentally exhibit useful results in terms of accuracy against exact, s
emiring-based justifications. In the experiments, we focus on tuples with
multiple generations of tuples in their lifelong lineage and analyze them
in terms of direct and distant lineage. The examples we present suggest a
high usefulness potential for the proposed approximate lineage methods an
d the further suggested enhancements. This especially holds for the column
-based vectors method which exhibits high precision and per-level recall.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture: 91832419086
UID:123se24012024100570
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20210601T113000
DTEND;TZID="Asia/Jerusalem":20210601T123000
DTSTAMP;TZID="Asia/Jerusalem":20210601T113000
FREEBUSY;FBTYPE=BUSY:20210601T113000/20210601T123000
SUMMARY;LANGUAGE=en-US:pixel-club talk by Prof. Adi Shamir (Weizmann Insti
tute of Science) about Pixel Club: A New Theory of Adversarial Examples in
Machine Learning at 2021-06-01 11:30:00
DESCRIPTION;LANGUAGE=en-US:The extreme fragility of deep neural networks w
hen presented with tiny perturbations in their inputs was independently di
scovered by several research groups in 2013. Due to their mysterious prope
rties and major security implications, these adversarial examples had been
studied extensively over the last eight years, but in spite of enormous e
ffort they remained a baffling phenomenon with no clear explanation. In pa
rticular, it was not clear why a tiny distance away from almost any cat im
age there are images which are recognized with a very high level of confid
ence as cars, planes, frogs, horses, or any other desired class, why the a
dversarial modification which turns a cat into a car does not look like a
car at all, and why a network which was adversarially trained with randoml
y permuted labels (so that it never saw any image which looks like a cat b
eing called a cat) still recognizes most cat images as cats.
The goal
of this talk is to introduce a new theory of adversarial examples, which
we call the Dimpled Manifold Model. It can easily explain in a simple and
intuitive way why they exist and why they have all the bizarre properties
mentioned above. In addition, it sheds new light on broader issues in mach
ine learning such as what happens to deep neural networks during regular a
nd during adversarial training. Experimental support for this theory, obta
ined jointly with Odelia Melamed and Oriel BenShmuel, will be presented an
d discussed in the last part of the talk.
Short bio:
Adi Shami
r is a professor at the Department of Mathematics and Computer Science at
the Weizmann Institute. He is well known for his fundamental contributions
in cryptography (Rivest–Shamir–Adleman (RSA) algorithm, Feige–Fiat–Shamir
identification scheme, differential cryptanalysis and more), and in other
computer science-related topics. Shamir is the recipient of the Turing Aw
ard (together with Adleman and Rivest), the Israel Mathematical Union Erd
ős Prize in Mathematics, and other awards. In 2019 he was elected as a Mem
ber of the American Philosophical Society.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture: 94741786518
UID:123se24012024100620
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20210603T163000
DTEND;TZID="Asia/Jerusalem":20210603T183000
DTSTAMP;TZID="Asia/Jerusalem":20210603T163000
FREEBUSY;FBTYPE=BUSY:20210603T163000/20210603T183000
SUMMARY;LANGUAGE=en-US:CSpecial Event about Online Lecture on the Way from
Taub to Google Japan at 2021-06-03 16:30:00
DESCRIPTION;LANGUAGE=en-US:You are invited to an online lecture by Sarai D
uak, a CS graduate and currently a Data Scientist Lead at Google Tokyo, Ja
pan, on simple solutions for business development problems for customers,
with the help of Data Science, on Thursday, June 3, 2021, 16:30.
Link
to the Zoom meeting will be sent upon pre-registration.
More details in the attached p
oster.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Event: Registration
UID:123se24012024100640
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20210607T113000
DTEND;TZID="Asia/Jerusalem":20210607T123000
DTSTAMP;TZID="Asia/Jerusalem":20210607T113000
FREEBUSY;FBTYPE=BUSY:20210607T113000/20210607T123000
SUMMARY;LANGUAGE=en-US:cggc talk by Rana Hanocka (Tel Aviv University) abo
ut CGGC Seminar: Neural 3D Reconstruction at 2021-06-07 11:30:00
DESCRIPTION;LANGUAGE=en-US:Neural networks have made exciting progress on
unstructured 3D geometric data; which is changing the way we fundamentally
approach problems in geometry processing. In this talk, I will discuss se
veral works which facilitate 3D reconstruction from several different dire
ctions, including consolidating point clouds, estimating a globally consis
tent point normal orientation, and reconstructing a surface mesh. Finally,
I will conclude with ongoing and future work in this direction, as well a
s other related areas.
The lecture will be recorded.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture: 91344952941
UID:123se24012024100720
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20210607T183000
DTEND;TZID="Asia/Jerusalem":20210607T203000
DTSTAMP;TZID="Asia/Jerusalem":20210607T183000
FREEBUSY;FBTYPE=BUSY:20210607T183000/20210607T203000
SUMMARY;LANGUAGE=en-US:CSpecial Event about Online Meeting with Microsoft
at 2021-06-07 18:30:00
DESCRIPTION;LANGUAGE=en-US:You are invited to an online meeting (TEAMS) wi
th Microsoft representatives and to hear from their students on the work e
xperience in the company and the combination of studies and careers, from
the managers and the recruitment team on job interviews, and more, on Mond
ay, June 7, 202, 18:30.
A link to the meeting will be sent upon pre-registration.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:TEAMS Event: Registration
UID:123se24012024100610
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20210608T100000
DTEND;TZID="Asia/Jerusalem":20210608T110000
DTSTAMP;TZID="Asia/Jerusalem":20210608T100000
FREEBUSY;FBTYPE=BUSY:20210608T100000/20210608T110000
SUMMARY;LANGUAGE=en-US:phd talk by Uri Alon about Machine Learning for Pro
gramming Language Processing at 2021-06-08 10:00:00
DESCRIPTION;LANGUAGE=en-US:This talk will focus on structural representati
ons and neural models of source code. I will present a language-agnostic a
pproach for structural language modeling (SLM) of code.
This general ap
proach obtains state-of-the-art results in a variety of tasks including co
de summarization, code captioning, code completion, name prediction, and e
dit completion, outperforming sequence models (such as textual Transformer
s and LSTMs) and models based on graph neural networks (GNNs).
Study
ing the reason why GNNs do poorly compared to SLM exposed a fundamental bo
ttleneck that results in a phenomenon that we call "over-squashing". This
bottleneck of GNNs provides a new perspective on a phenomenon that was obs
erved for years and still affects existing GNN-based models. I will explai
n the GNN bottleneck problem and demonstrate its practical implications.
The talk summarizes my PhD dissertation, which has been presented in b
oth machine learning and programming language conferences.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture: 3739170675
UID:123se24012024100680
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20210608T113000
DTEND;TZID="Asia/Jerusalem":20210608T123000
DTSTAMP;TZID="Asia/Jerusalem":20210608T113000
FREEBUSY;FBTYPE=BUSY:20210608T113000/20210608T123000
SUMMARY;LANGUAGE=en-US:pixel-club talk by Prof. Shimon Ullman (Weizmann In
stitute of Science) about Pixel Club: Scene Understanding by Iterative Bot
tom-up top-down Processing at 2021-06-08 11:30:00
DESCRIPTION;LANGUAGE=en-US:Scene understanding requires the extraction and
representation of scene components together with their individual propert
ies, as well relations and interactions between them. In current computer
vision, there has been considerable progress in recognizing scene componen
ts (people, objects, parts), but the problem of recovering scene structure
is still largely open.
I will describe a model that performs scene
interpretation by an iterative process, combining bottom-up and top-down
networks, interacting through a symmetric bi-directional communication bet
ween them. The model extracts and recognizes scene components with their s
elected properties and relations, and uses them to describe and understand
the image.
Short Bio:
Shimon Ullman is the Samy and Ruth Cohn Pro
fessor of Computer Science at the Weizmann Institute of Science, and the h
ead of the Weizmann Institute AI Center. Prior to this position, he was a
Professor at the Brain and Cognitive Science and the AI Laboratory at MIT.
His areas of research combine computer and human vision, human cognition,
and brain modeling.
He obtained his B.Sc. in Mathematics, Physics an
d Biology, at the Hebrew University of Jerusalem, and Ph.D. in Electrical
Engineering and Computer Sciences, at the Artificial Intelligence Laborato
ry in the Massachusetts Institute of Technology. He is a recipient of the
2008 David. E. Rumelhart Prize in human cognition, the 2014 Emet Prize for
Art, Science and Culture, the 2015 Israel Prize in Computer Science, and
the 2019 IEEE Azriel Rosenfeld Award for lifetime achievement in computer
vision. He is a member of the Israeli Academy of Sciences and Humanities,
and the American Academy of Arts and Sciences.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture: 99725686717
UID:123se24012024100710
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20210609T133000
DTEND;TZID="Asia/Jerusalem":20210609T143000
DTSTAMP;TZID="Asia/Jerusalem":20210609T133000
FREEBUSY;FBTYPE=BUSY:20210609T133000/20210609T143000
SUMMARY;LANGUAGE=en-US:msc talk by Shai Ben-Assayag about Scaling Up Board
Games with AlphaZero and Graph Neural Networks at 2021-06-09 13:30:00
DESCRIPTION;LANGUAGE=en-US:Playing board games is considered a major chall
enge for both humans and AI researchers. Because some complicated board ga
mes are quite hard to learn, humans usually begin with playing on smaller
boards and incrementally advance to master larger board strategies. Most n
eural network frameworks that are currently tasked with playing board game
s neither perform such incremental learning nor possess capabilities to au
tomatically scale up. In this work, we look at the board as a graph and co
mbine a graph neural network architecture inside the AlphaZero framework,
along with some other innovative improvements. Our ScalableAlphaZero is ca
pable of learning to play incrementally on small boards, and advancing to
play on large ones. Our model can be trained quickly to play different cha
llenging board games on multiple board sizes, without using any domain kno
wledge. We demonstrate the effectiveness of ScalableAlphaZero and show, fo
r example, that by training it for only three days on small Othello boards
, it can defeat the AlphaZero model on a large board, which was trained to
play the large board for 30 days.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture: 4222318274
UID:123se24012024100650
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20210615T113000
DTEND;TZID="Asia/Jerusalem":20210615T123000
DTSTAMP;TZID="Asia/Jerusalem":20210615T113000
FREEBUSY;FBTYPE=BUSY:20210615T113000/20210615T123000
SUMMARY;LANGUAGE=en-US:pixel-club talk by Tammy Riklin Raviv (Ben-Gurion U
niversity) about Pixel Club: Subsampled Brain MRI Reconstructionby Genera
tive Adversarial Neural Networks at 2021-06-15 11:30:00
DESCRIPTION;LANGUAGE=en-US:A main challenge in magnetic resonance imaging
(MRI) is speeding up scan time. Beyond improving patient experience and re
ducing operational costs, faster scans are essential for time-sensitive im
aging, such as fetal, cardiac, or functional MRI, where temporal resolutio
n is important and target movement is unavoidable, yet must be reduced. Cu
rrent MRI acquisition methods speed up scan time at the expense of lower s
patial resolution and costlier hardware. We introduce a practical, softwar
e-only framework, based on deep learning, for accelerating MRI acquisition
, while maintaining anatomically meaningful imaging. This is accomplished
by MRI subsampling followed by estimating the missing k-space samples via
generative adversarial neural networks. A generator-discriminator interpla
y enables the introduction of an adversarial cost in addition to fidelity
and image-quality losses used for optimizing the reconstruction.
Prom
ising reconstruction results are obtained from feasible sampling patterns
of up to a fivefold acceleration of diverse brain MRIs, from a large publi
cly available dataset of healthy adult scans as well as multimodal acquisi
tions of multiple sclerosis patients and dynamic contrast-enhanced MRI (DC
E-MRI) sequences of stroke and tumor patients. Clinical usability of the r
econstructed MRI scans is assessed by performing either lesion or healthy
tissue segmentation and comparing the results to those obtained by using t
he original, fully sampled images. Reconstruction quality and usability of
the DCE-MRIsequences is demonstrated by calculating the pharmacokinetic (
PK) parameters. The proposed MRI reconstruction approach is shown to outpe
rform state-of-the-art methods for all datasets tested in terms of the pea
k signal-to-noise ratio (PSNR), the structural similarity index (SSIM), as
well as either the mean squared error (MSE) with respect to the PK parame
ters, calculated for the fully sampled DCE-MRI sequences, or the segmentat
ion compatibility, measured in terms of Dice scores and Hausdorff distance
. The code is available on GitHub.
Short Bio:
Dr. Tammy Riklin Rav
iv is a faculty member at the School of Electrical and Computer Engineerin
g of Ben-Gurion University (BGU) since 2012. In the recent years her resea
rch group focuses on the development of deep learning algorithms for addre
ssing a variety of Biomedical Imaging analysis problems. She holds a B.Sc.
in Physics (since 1993) and an M.Sc. in Computer Science (both magna cum
laude) from the Hebrew University of Jerusalem. In 2008, she received a Ph
.D. from the School of Electrical Engineering of Tel-Aviv University (2008
). In the years 2008-2012 she was a post-doctorate associate and a researc
h fellow at the Computer Science and Artificial Intelligence Laboratory (C
SAIL) of MIT, Harvard Medical School and the Broad Institute of MIT and Ha
rvard. Dr. Riklin Raviv serves as an Associate Editor at the IEEE Transact
ion on Medical Imaging (IEEE TMI) journal and as a handling editor in Neur
oImage. She is also a Technical Committee (TC) member at the IEEE Bio Imag
ing and Signal Processing (BISP) Committee.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture: 94176564806
UID:123se24012024100770
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20210616T110000
DTEND;TZID="Asia/Jerusalem":20210616T120000
DTSTAMP;TZID="Asia/Jerusalem":20210616T110000
FREEBUSY;FBTYPE=BUSY:20210616T110000/20210616T120000
SUMMARY;LANGUAGE=en-US:phd talk by Tao Hong about Numerical Optimization a
nd Multigrid Computational Methods with Applications at 2021-06-16 11:00:0
0
DESCRIPTION;LANGUAGE=en-US:Work 1: we introduce a way to adapt Nesterov's
well-known scheme to accelerating stationary iterative solvers for linear
systems. Compared with classical Krylov subspace acceleration methods, the
proposed scheme requires more iterations, but it is trivial to implement
and retains essentially the same computational cost as the unaccelerated m
ethod. An explicit formula for a fixed optimal parameter is derived in the
case where the stationary iteration matrix has only real eigenvalues, bas
ed only on the smallest and largest eigenvalues. The fixed parameter, and
corresponding convergence factor, are shown to maintain their optimality w
hen the iteration matrix also has complex eigenvalues that are contained w
ithin an explicitly defined disk in the complex plane. A comparison to Che
byshev acceleration based on the same information of the smallest and larg
est real eigenvalues (dubbed Restricted Information Chebyshev acceleration
) demonstrates that Nesterov's scheme is more robust in the sense that it
remains optimal over a larger domain when the iteration matrix does have s
ome complex eigenvalues.
Work 2: we introduce a general framework cal
led weighted proximal methods (WPMs) for regularization by denoising (RED)
model which uses abstract image demonising algorithms to build the prior.
In work, we first show that two recently introduced RED solvers (using t
he fixed point and accelerated proximal gradient methods) are particular c
ases of WPMs. Then we show by numerical experiments that slightly more sop
histicated variants of WPM can lead to reduced run times for RED by requir
ing a significantly smaller number of calls to the denoiser.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture: 96123577236
UID:123se24012024100690
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20210616T113000
DTEND;TZID="Asia/Jerusalem":20210616T123000
DTSTAMP;TZID="Asia/Jerusalem":20210616T113000
FREEBUSY;FBTYPE=BUSY:20210616T113000/20210616T123000
SUMMARY;LANGUAGE=en-US:cggc talk by Olga Diamanti (TU Graz, Institute for
Geometry) about CGGC Seminar: Discrete Willmore Surfaces at 2021-06-16 11:
30:00
DESCRIPTION;LANGUAGE=en-US:This talk will be about the problem of discrete
constrained Willmore surfaces: discrete surfaces that have minimal total
squared mean curvature while also being discretely conformally equivalent
to a given input surface. The Willmore energy is a bending energy, used to
model elastic behavior and measure surface smoothness. Adding the conform
ality constraint turns the problem into a natural extension, in 2D, of cla
ssical elastic spline modeling in 1D. This not only makes the use of Willm
ore functional more practical for a geometric modeling setting, but also l
eads to more interesting, visually appealing surfaces with rich geometric
features. In this talk, I will discuss both theoretical contributions, as
well as a practical and efficient algorithm to solve this numerically chal
lenging problem.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture: 91344952941
UID:123se24012024100750
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20210616T120000
DTEND;TZID="Asia/Jerusalem":20210616T130000
DTSTAMP;TZID="Asia/Jerusalem":20210616T120000
FREEBUSY;FBTYPE=BUSY:20210616T120000/20210616T130000
SUMMARY;LANGUAGE=en-US:msc talk by Adar Amir about DLACEP: A Deep-Learning
Based Framework for Approximate Complex Event Processing at 2021-06-16 12
:00:00
DESCRIPTION;LANGUAGE=en-US:Complex event processing (CEP) is employed to d
etect user-specified patterns of events in data streams. CEP mechanisms op
erate by maintaining all sets of events that can potentially be composed i
nto a pattern match. This approach can be wasteful when many of the sets d
o not participate in an actual match and are therefore discarded.
We
present DLACEP, a novel framework that fuses deep learning with CEP to eff
iciently extract complex pattern matches from streams. To the best of our
knowledge, this is the first time deep learning is employed to detect even
ts constituting a pattern match in the realm of CEP. To assess our approac
h, we performed extensive empirical testing on various scenarios with both
synthetic and real-world data. We showcase examples in which our method a
chieves an increase in throughput of up to three orders of magnitude compa
red to solely employing CEP, while only suffering a minor loss in the numb
er of detected matches.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture: 99646066466
UID:123se24012024100740
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20210616T123000
DTEND;TZID="Asia/Jerusalem":20210616T143000
DTSTAMP;TZID="Asia/Jerusalem":20210616T123000
FREEBUSY;FBTYPE=BUSY:20210616T123000/20210616T143000
SUMMARY;LANGUAGE=en-US:CSpecial Event about Best Project Contest - The Fin
als at 2021-06-16 12:30:00
DESCRIPTION;LANGUAGE=en-US:You are invited to the finals event of the Best
Project Contest, to be held on Wednesday, June 16, 2021, starting at 12:3
0 and at 14:00 announcing and awarding the winners, at the CS Taub Lobby.
The event will take place in the format of a project fair, and in acc
ordance with the guidelines of the green pass instructions.
You are
all invited to come and meet the best researchers and researches!
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:CS Taub Lobby
UID:123se24012024100580
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20210616T180000
DTEND;TZID="Asia/Jerusalem":20210616T200000
DTSTAMP;TZID="Asia/Jerusalem":20210616T180000
FREEBUSY;FBTYPE=BUSY:20210616T180000/20210616T200000
SUMMARY;LANGUAGE=en-US:CSpecial Event about Meetup by RAFAEL at 2021-06-16
18:00:00
DESCRIPTION;LANGUAGE=en-US:Rafael will hold a Meetup meeting with the part
icipation of Gidi Weiss, VP of Marketing and Business Development in the d
ivision, who will talk about the most advanced security technologies in th
e world.
The meeting will take place on Wednesday, May 26, 2021, at t
he Nola Socks Pub, Haifa, and participation requires pre-registration.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Nola Socks Pub, Haifa
UID:123se24012024100590
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20210621T163000
DTEND;TZID="Asia/Jerusalem":20210621T173000
DTSTAMP;TZID="Asia/Jerusalem":20210621T163000
FREEBUSY;FBTYPE=BUSY:20210621T163000/20210621T173000
SUMMARY;LANGUAGE=en-US:phd talk by Ariel Kulik about Analysis of Two-varia
ble Recurrence Relations with Application to Parameterized Approximations
at 2021-06-21 16:30:00
DESCRIPTION;LANGUAGE=en-US:We introduce randomized branching as a tool for
parameterized approximation and develop the mathematical machinery for it
s analysis. Our algorithms substantially improve the best known running ti
mes of parameterized approximation algorithms for Vertex Cover and $3$-Hit
ting Set for a wide range of approximation ratios.
The running times
of our algorithms are derived from an asymptotic analysis of a broad class
of two-variable recurrence relations. Our main theorem gives a simple for
mula for this asymptotics. The formula can be efficiently calculated by s
olving a simple numerical optimization problem, and provides the mathemati
cal insight required for the algorithm design. To this end, we show an equ
ivalence between the recurrence and a stochastic process. We analyze this
process using the Method of types, by introducing an adaptation of Sanov
's theorem to our setting.
The talk will include a survey of additio
nal results obtained during my PhD studies.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture: 99623903736
UID:123se24012024100700
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20210622T113000
DTEND;TZID="Asia/Jerusalem":20210622T123000
DTSTAMP;TZID="Asia/Jerusalem":20210622T113000
FREEBUSY;FBTYPE=BUSY:20210622T113000/20210622T123000
SUMMARY;LANGUAGE=en-US:pixel-club talk by Boaz Nadler (Weizmann Institute
of Science) about Pixel Club: The Trimmed Lasso: Sparse Recovery Guarantee
s and Practical Optimization at 2021-06-22 11:30:00
DESCRIPTION;LANGUAGE=en-US:Consider the sparse approximation or best subse
t selection problem:
Given a vector y and a matrix A, find a k-sparse ve
ctor x that minimizes the residual ||Ax-y||.
This sparse linear regre
ssion problem, and related variants, plays a key role in high dimensional
statistics, compressed sensing, machine learning and more.
In this ta
lk we focus on the trimmed lasso penalty, defined as the L_1 norm of x min
us the L_1 norm of its top k entries in absolute value. We advocate using
this penalty by deriving sparse recovery guarantees for it, and by presen
ting a practical approach to optimize it. Our computational approach is ba
sed on the generalized soft-min penalty, a smooth surrogate that takes int
o account all possible k-sparse patterns.
We derive a polynomial time
algorithm to compute it, which in turn yields a novel method for the best
subset selection problem. Numerical simulations illustrate its competitiv
e performance compared to current state of the art.
Joint work with
Tal Amir and Ronen Basri.
Short Bio:
Boaz Nadler is a professor a
t the department of computer science at the Weizmann Institute of Science.
His research interests include mathematical statistics, machine learning,
signal and image processing.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture: 95693258864
UID:123se24012024100820
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20210622T123000
DTEND;TZID="Asia/Jerusalem":20210622T133000
DTSTAMP;TZID="Asia/Jerusalem":20210622T123000
FREEBUSY;FBTYPE=BUSY:20210622T123000/20210622T133000
SUMMARY;LANGUAGE=en-US:msc talk by Entony Lekhtman about Domain Adaptation
with Category Shift, an Application to Aspect Extraction at 2021-06-22 12
:30:00
DESCRIPTION;LANGUAGE=en-US:The rise of pre-trained language models has yie
lded substantial progress in the vast majority of Natural Language Process
ing (NLP) tasks. However, a generic approach towards the pre-training proc
edure can naturally be sub-optimal in some cases. Particularly, fine-tunin
g a pre-trained language model on a source domain and then applying it to
a different target domain, results in a sharp performance decline of the
eventual classifier for many source-target domain pairs. Moreover, in some
NLP tasks, the output categories substantially differ between domains, ma
king adaptation even more challenging. This, for example, happens in the t
ask of aspect extraction, where the aspects of interest of reviews of, e.g
., restaurants or electronic devices may be very different. In this work
we present a new fine-tuning scheme for BERT, which aims to address the ab
ove challenges.
We name this scheme DILBERT: Domain Invariant Learni
ng with BERT, and customize it for aspect extraction in the unsupervised
domain adaptation setting. DILBERT harnesses the categorical information o
f both the source and the target domains to guide the pre-training process
towards a more domain and category invariant representation, thus closing
the gap between the domains. We show that DILBERT yields substantial impr
ovements over state-of-the-art baselines while using a fraction of the unl
abeled data, particularly in more challenging domain adaptation setups.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture: 92960384508
UID:123se24012024100790
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20210623T100000
DTEND;TZID="Asia/Jerusalem":20210623T170000
DTSTAMP;TZID="Asia/Jerusalem":20210623T100000
FREEBUSY;FBTYPE=BUSY:20210623T100000/20210623T170000
SUMMARY;LANGUAGE=en-US:CSpecial Event about Recruitment Day by Vayyar at 2
021-06-23 10:00:00
DESCRIPTION;LANGUAGE=en-US:Vayyar representatives will visit CS to demonst
rate their Radar-based technological solutions, on Wednesday, June 23, 202
1, between 10:00-17:00, at the CS Taub Lobby.
More details in the att
ached poster.
You are all invited!
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:CS Taub Lobby
UID:123se24012024100780
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20210623T113000
DTEND;TZID="Asia/Jerusalem":20210623T123000
DTSTAMP;TZID="Asia/Jerusalem":20210623T113000
FREEBUSY;FBTYPE=BUSY:20210623T113000/20210623T123000
SUMMARY;LANGUAGE=en-US:ceClub talk by Meni Orenbach (EE, Technion) about c
eClub: Operating Systems Abstractions for Trusted Execution Environments a
t 2021-06-23 11:30:00
DESCRIPTION;LANGUAGE=en-US:Trusted execution environments such as secure e
nclaves are now available in several popular CPUs, and supported in public
clouds. Enclaves can be used to efficiently shield applications against p
rivileged adversaries, and secure sensitive data processed by them through
strong isolation backed by the hardware. Yet, enclaves are not a silver b
ullet: they are vulnerable to unique side-channel attacks, they exhibit po
or performance when system calls are invoked and when page faults occur, t
hey lack a secure variant of software abstractions such as page fault hand
lers, and finally, the hardware does not protect against Iago attacks.
I
n our work, we tackle the aforementioned shortcomings of existing enclaves
with system abstractions, practical hardware modifications, and tools to
support them. In this talk, we provide a high-level overview of our approa
ch followed by presenting TEEProtect, a framework for thwarting Iago attac
ks.
* Ph.D. student under supervision of Prof. Mark Silberstein.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture: 95666603201
UID:123se24012024100760
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20210627T113000
DTEND;TZID="Asia/Jerusalem":20210627T123000
DTSTAMP;TZID="Asia/Jerusalem":20210627T113000
FREEBUSY;FBTYPE=BUSY:20210627T113000/20210627T123000
SUMMARY;LANGUAGE=en-US:TDC talk by Shimon Biton (IE, Technion) about Distr
ibuted Computing Seminar: A Fully Adaptive Self-Stabilizing Transformer fo
r LCL Problems at 2021-06-27 11:30:00
DESCRIPTION;LANGUAGE=en-US:The first generic self-stabilizing transformer
for local problems in a constrained bandwidth model is introduced. This tr
ansformer can be applied to a wide class of locally checkable labeling (LC
L) problems, converting a given fault free synchronous algorithm that sati
sfies certain conditions into a self-stabilizing synchronous algorithm for
the same problem.
The resulting self-stabilizing algorithms are anon
ymous, size-uniform, and \emph{fully adaptive} in the sense that their tim
e complexity is bounded as a function of the number $k$ of nodes that suff
ered faults (possibly at different times) since the last legal configurati
on.
Specifically, for graphs whose degrees are up-bounded by $\Delta$
, the algorithms produced by the transformer stabilize in time proportiona
l to $\log (k + \Delta)$ in expectation, independently of the number of no
des in the graph (in some cases, the dependency on $\Delta$ can also be om
itted).
As such, the transformer is applicable also for infinite grap
hs (with degree bound $\Delta$). Another appealing feature of the transfor
mer is its small message size overhead.
The transformer is applied to
known algorithms (or simple variants thereof) for some classic LCL proble
ms, producing the first anonymous size-uniform self-stabilizing algorithms
for these problems that are provably fully adaptive.
Fom a technical
point of view, the transformer's key design feature is a novel probabilis
tic tool that allows different nodes to act in synchrony even though their
clocks may have been adversarially manipulated.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture: 99794260392 and Bloomfield 152 (Hybrid manner)
UID:123se24012024100830
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20210628T113000
DTEND;TZID="Asia/Jerusalem":20210628T123000
DTSTAMP;TZID="Asia/Jerusalem":20210628T113000
FREEBUSY;FBTYPE=BUSY:20210628T113000/20210628T123000
SUMMARY;LANGUAGE=en-US:cggc talk by Yotam Gingold (George Mason University
) about CGGC Seminar: Hyperspectral Inverse Skinning at 2021-06-28 11:30:0
0
DESCRIPTION;LANGUAGE=en-US:In example-based inverse linear blend skinning
(LBS), a collection of poses (e.g., animation frames) are given, and the g
oal is finding skinning weights and transformation matrices that closely r
eproduce the input. These poses may come from physical simulation, direct
mesh editing, motion capture, or another deformation rig. We describe a re
-formulation of inverse skinning as a problem in high-dimensional Euclidea
n space. The transformation matrices applied to a vertex across all poses
can be thought of as a point in high dimensions. We cast the inverse LBS p
roblem as one of finding a tight-fitting simplex around these points (a we
ll-studied problem in hyperspectral imaging). Although we do not observe t
ransformation matrices directly, the 3D position of a vertex across all of
its poses defines an affine subspace, or flat. We solve a “closest flat”
optimization problem to find points on these flats, and then compute a min
imum-volume enclosing simplex whose vertices are the transformation matric
es and whose barycentric coordinates are the skinning weights. We are able
to create LBS rigs with state-of-the-art reconstruction error, and state-
of-the-art compression ratios for mesh animation sequences. Our solution d
oes not consider weight sparsity or the rigidity of recovered transformati
ons. We include observations and insights into the closest flat problem. I
ts ideal solution, and optimal LBS reconstruction error, remain an open pr
oblem.
Bio: Yotam Gingold is an Associate Professor in the computer s
cience department at George Mason University. He directs the Creativity an
d Graphics Lab (CraGL), whose mission is to solve challenging visual, geom
etry, and design problems and pursue foundational research into human crea
tivity. His work has been supported by the National Science Foundation (in
cluding a CAREER award), Google, and Adobe. His research has been incorpor
ated into Adobe Illustrator as the Puppet Warp tool. Previously he was a p
ost-doctoral researcher in the computer science departments of Columbia Un
iversity, Rutgers University, Tel-Aviv University, and Herzliya IDC. Yotam
earned his Ph.D. in Computer Science from New York University in 2009, wh
ere he was awarded the Janet Fabri Prize for most outstanding dissertation
.
The talk will be given in hybrid mode, in room 337.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture: 91344952941
UID:123se24012024100850
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20210629T123000
DTEND;TZID="Asia/Jerusalem":20210629T143000
DTSTAMP;TZID="Asia/Jerusalem":20210629T123000
FREEBUSY;FBTYPE=BUSY:20210629T123000/20210629T143000
SUMMARY;LANGUAGE=en-US:CSpecial Event about Project Fair in IoT, Software,
Android Apps, AI, Cyber, Computer Security, and Networks at 2021-06-29 1
2:30:00
DESCRIPTION;LANGUAGE=en-US:CS Labs: Systems and Software Deve
lopment Laboratory (SSDL), Cyber and Computer Security Laboratory (CYBER),
The Laboratory for Computer Communication and Networking (LCCN) in
vite you to visit the Spring Project Fair in IoT, Softw
are, Android Apps, AI, Cyber, Computer Security,
and Networks, including demos and presentations
by 40 undergraduate teams who will answer your questions on their researc
h.
The event will be held on Tuesday, June 29, 2021, at 1
Mor
e details in UG secretariat.
Welcome to CS
, good luck and a successful academic year!
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:CS Taub Lobby and Taub Auditorium 1
UID:123se24012024101130
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20211025T133000
DTEND;TZID="Asia/Jerusalem":20211025T143000
DTSTAMP;TZID="Asia/Jerusalem":20211025T133000
FREEBUSY;FBTYPE=BUSY:20211025T133000/20211025T143000
SUMMARY;LANGUAGE=en-US:msc talk by Majd Khalil about The Complexity of the
Shapley Value for Path Queries over Graphs at 2021-10-25 13:30:00
DESCRIPTION;LANGUAGE=en-US:A path query extracts from the input graph the
pairs of vertices that constitute the endpoints of matching paths, that is
, paths such that the word obtained from the edge labels belongs to a spe
cified language. We study the computational complexity of measuring the co
ntribution of edges and vertices to an answer of a path query.
For th
at, we adopt the traditional Shapley value from cooperative game theory. T
his value has recently been suggested and studied as a standard contributi
on measure for database queries, machine-learning classifiers, and so on.
We start by showing that the exact Shapley value is almost always har
d to compute. For example, we show that (under conventional complexity ass
umptions) the Shapley value of an edge can be computed in polynomial time
if and only if the path language has only words of length at most two.
On the other hand, it is straightforward to obtain an efficient scheme
(FPRAS) for an additive approximation. Yet, a multiplicative approximation
is harder to obtain. We establish that in the case of a regular language,
a multiplicative approximation of the Shapley value of an edge can be com
puted in polynomial time if and only if the path language is finite.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture: 95634964056
UID:123se24012024101140
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20211026T143000
DTEND;TZID="Asia/Jerusalem":20211026T153000
DTSTAMP;TZID="Asia/Jerusalem":20211026T143000
FREEBUSY;FBTYPE=BUSY:20211026T143000/20211026T153000
SUMMARY;LANGUAGE=en-US:msc talk by Evgenii Zheltonozhskii about Reducing
Supervision in Visual Recognition Tasks at 2021-10-26 14:30:00
DESCRIPTION;LANGUAGE=en-US:While deep neural networks (DNNs) have shown tr
emendous success across various computer vision tasks, including image cla
ssification, object detection, and semantic segmentation, requirements for
a large number of high-quality labels obstruct the adoption of DNNs in re
al-life problems. Lately, researchers have proposed multiple approaches fo
r reducing requirements to the amount or quality of these labels or even w
orking in a fully unsupervised way. In a series of works, we study differe
nt approaches to supervision reduction in visual recognition tasks: self-s
upervised learning, learning with noisy labels, and semi-supervised learni
ng. For self-supervised learning, we show that dimensionality reduction fo
llowed by simple k-nearest neighbors clustering is a very strong baseline
for fully unsupervised large-scale classification (ImageNet). Additionally
, we present a learning with noisy labels framework comprising two stages:
self-supervised pre-training and robust fine-tuning. The framework, dubbe
d "Contrast to Divide" (C2D), significantly outperforms prior art on synth
etic and real-life noise, showing state-of-the-art performance with differ
ent methods and pre-training approaches. Furthermore, since self-supervise
d pre-training is unaffected by label noise, C2D is especially efficient i
n a high noise regime. Finally, for semi-supervised learning, we propose a
simple weighting scheme that reduces confirmation bias among unlabeled sa
mples and, as a result, outperforms existing methods on different datasets
and a wide range of labeled sample fractions. The talk will be given in E
nglish.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture: 5447249519 and Taub 014
UID:123se24012024101120
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20211027T173000
DTEND;TZID="Asia/Jerusalem":20211027T193000
DTSTAMP;TZID="Asia/Jerusalem":20211027T173000
FREEBUSY;FBTYPE=BUSY:20211027T173000/20211027T193000
SUMMARY;LANGUAGE=en-US:CSpecial Talk about Recruitment Day by NVIDIA at 20
21-10-27 17:30:00
DESCRIPTION;LANGUAGE=en-US:CS students are invited to Recruitment Day by N
VIDIA to learn about its activity, to hear lectures on Software, Fi
rmware, Architectura, Chip Design and to meet its engin
eers who will answer you questions and tell you about their employment opt
ions.
The event will be held via Teams on Wednesday, October 27, 2021, between 17:20-19:00 and
requires pre-registration.
During the event, a logic puzzle bearing a
cash prize for a vacation in Israel or abroad will be published, and in or
der to participate in the solution, you must register in ad
vance and send the answers by email: bbentolila@nvidia.co
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Teams Event: shorturl.at/hnAST
UID:123se24012024101180
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20211102T170000
DTEND;TZID="Asia/Jerusalem":20211102T180000
DTSTAMP;TZID="Asia/Jerusalem":20211102T170000
FREEBUSY;FBTYPE=BUSY:20211102T170000/20211102T180000
SUMMARY;LANGUAGE=en-US:msc talk by Koral Chapnik about DARLING: Data-Aware
Load Shedding in Complex Event Processing Systems at 2021-11-02 17:00:00
DESCRIPTION;LANGUAGE=en-US:Complex event processing (CEP) is widely employ
ed to detect user-defined combinations, or patterns, of events in massive
streams of incoming data. Numerous applications such as healthcare, fraud
detection, and more, use CEP technologies to capture critical alerts, thre
ats, or vital notifications. This requires that the technology meet real-t
ime detection constraints. Multiple optimization techniques have been deve
loped to minimize the processing time for CEP, including parallelization t
echniques, pattern rewriting, and more. However, these techniques may not
suffice or may not be applicable when an unpredictable peak in the input e
vent stream exceeds the system capacity. In such cases, one immediate poss
ible solution is to drop some of the load in a technique known as load she
dding.
We present a novel load shedding mechanism for real-time compl
ex event processing. Our approach uses statistics that are gathered to det
ect overload. The solution makes data-driven load shedding decisions to dr
op the less important events such that we preserve a given latency bound w
hile minimizing the degradation in the quality of results. An extensive ex
perimental evaluation on a broad set of real-life patterns and datasets de
monstrates the superiority of our approach over the state-of-the-art techn
iques.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture: 5602945484
UID:123se24012024101150
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20211102T183000
DTEND;TZID="Asia/Jerusalem":20211102T203000
DTSTAMP;TZID="Asia/Jerusalem":20211102T183000
FREEBUSY;FBTYPE=BUSY:20211102T183000/20211102T203000
SUMMARY;LANGUAGE=en-US:CSpecial Event about How Do You Turn a Degree into
a Career? at 2021-11-02 18:30:00
DESCRIPTION;LANGUAGE=en-US:How Do You Turn a Degree into a Career?
The development of computing systems able to address our ever-increasing needs, especially as we reach the end of CMOS transistor scaling, requires truly novel methods of computing. My researc h draws inspiration from biology, rethinks the digital/analog boundary, a nd challenges conventional wisdom, which typically guides how we perform c omputation, by reimagining the role of time. In this talk, I first introdu ce a computational temporal logic that sets the foundation for temporal co mputing. Second, I demonstrate how this foundation opens up unique ways in which we can work with sensors and design machine learning systems. Third , I describe how temporal operators provide answers to several long-lastin g problems in computing with emerging devices. Finally, I touch upon futur e work with themes ranging from in-sensor online learning to hybrid quantu m-classical computing and formally verifiable hardware.
Bio: Ge orge Tzimpragos is a Ph.D. candidate in Computer Science at UC Santa Barba ra and a research affiliate at Lawrence Berkeley National Laboratory. His research explores how computer architecture concepts, along with a deeper understanding of the nature of computation and devices, can be used to dev elop new paradigms and cross-stack solutions for emerging applications and technologies. His work has been published in top conferences and journals and rewarded with an ASPLOS best paper award, an IEEE Micro top pick, a C ACM research highlight, and an invited oral presentation at EUCAS. He is a lso the creator of https://thejjunction.org, a platform for the exchange o f ideas and resources on superconducting computing. His personal website i s https://georgetzimpragos.com.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il LOCATION:Zoom Lecture: 96743325005 UID:123se24012024101870 END:VEVENT BEGIN:VEVENT DTSTART;TZID="Asia/Jerusalem":20220116T103000 DTEND;TZID="Asia/Jerusalem":20220116T113000 DTSTAMP;TZID="Asia/Jerusalem":20220116T103000 FREEBUSY;FBTYPE=BUSY:20220116T103000/20220116T113000 SUMMARY;LANGUAGE=en-US:msc talk by Alon Dankner about Securing ICS Protoco ls at 2022-01-16 10:30:00 DESCRIPTION;LANGUAGE=en-US:Industrial Control Systems (ICSs), also known a s Operation Technology (OT) systems, are distributed computerized systems designed to manage, monitor and control industrial processes. They are w idely used in critical infrastructures, such as power plants and water sup ply, whose continuous operation is of major importance to modern life. F ollowing the well-known Stuxnet attack on OT systems, a large investment i n OT security was started. Though their cyber protection is crucial, the y did not yet reach the same level of cyber protection maturity as Informa tion Technology (IT) systems. In this research, we make two complemen tary contributions. The first is to show that the existing systems are vul nerable by demonstrating novel attacks against ICS systems. The second i s to propose a secure architecture and protocol for ICS systems in large o rganizations. Our protocol is carefully designed to defend against attacks and support the unique requirements of these environemnts (e.g., central management of devices and support for legacy devices). ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il LOCATION:Zoom Lecture: 8899993884 UID:123se24012024101830 END:VEVENT BEGIN:VEVENT DTSTART;TZID="Asia/Jerusalem":20220116T150000 DTEND;TZID="Asia/Jerusalem":20220116T160000 DTSTAMP;TZID="Asia/Jerusalem":20220116T150000 FREEBUSY;FBTYPE=BUSY:20220116T150000/20220116T160000 SUMMARY;LANGUAGE=en-US:msc talk by Tom Brand about Bribery attack on Nakam oto Consensus Proof of Stake Protocols at 2022-01-16 15:00:00 DESCRIPTION;LANGUAGE=en-US:Bitcoin was introduced to the world in 2009 wit h Proof of Work (PoW) Leader Election as one of its novel building blocks. Since then, much criticism has been made of its high energy consumption. Proof of Stake protocols aims at replacing PoW protocols as a much more ef ficient version while still maintaining its security properties under the Honest Majority model. In our work, we show a bribery attack under th e Rational Majority model, which breaks the persistence security property of the underlying protocol. Furthermore, we show how this attack can be co nducted on a specific protocol - Ouroboros Praos. Finally, we propose a so lution for making it as secure as PoW protocols. ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il LOCATION:Zoom Lecture: 7228552597 UID:123se24012024101820 END:VEVENT BEGIN:VEVENT DTSTART;TZID="Asia/Jerusalem":20220118T113000 DTEND;TZID="Asia/Jerusalem":20220118T123000 DTSTAMP;TZID="Asia/Jerusalem":20220118T113000 FREEBUSY;FBTYPE=BUSY:20220118T113000/20220118T123000 SUMMARY;LANGUAGE=en-US:pixel-club talk by Tavi Halperin (The Hebrew Univer sity of Jerusalem) about Pixel Club: Endless Loops: Detecting and Animatin g Periodic Patterns in Still Images at 2022-01-18 11:30:00 DESCRIPTION;LANGUAGE=en-US:We present an algorithm for producing a seamles s animated loop from a single image. The algorithm detects periodic struct ures, such as the windows of a building or the steps of a staircase, and g enerates a non-trivial displacement vector field that maps each segment of the structure onto a neighboring segment along a user- or auto-selected m ain direction of motion. This displacement field is used, together with su itable temporal and spatial smoothing, to warp the image and produce the f rames of a continuous animation loop. Our cinemagraphs are created in unde r a second on a mobile device. Over 140,000 users downloaded our app and e xported over 350,000 cinemagraphs. Moreover, we conducted two user studies that show that users prefer our method for creating surreal and structure d cinemagraphs compared to more manual approaches and compared to previous methods. Bio: Tavi Halperin is a Researcher at Lightricks. He rece ived his Ph.D. in Computer Science from the Hebrew University of Jerusalem in 2019, advised by Prof. Shmuel Peleg. ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il LOCATION:Zoom Lecture: https://technion.zoom.us/my/chaimbaskin UID:123se24012024101880 END:VEVENT BEGIN:VEVENT DTSTART;TZID="Asia/Jerusalem":20220118T133000 DTEND;TZID="Asia/Jerusalem":20220118T143000 DTSTAMP;TZID="Asia/Jerusalem":20220118T133000 FREEBUSY;FBTYPE=BUSY:20220118T133000/20220118T143000 SUMMARY;LANGUAGE=en-US:msc talk by Amir Livne about How to Avoid Depth Rec onstruction in 3D Vision Tasks: Do We Need Depth in State-Of-The-Art Face Authentication? at 2022-01-18 13:30:00 DESCRIPTION;LANGUAGE=en-US:Face recognition systems are frequently used in a variety of security applications in our daily lives. Some methods are d esigned to utilize geometric information extracted from depth sensors to o vercome single-image-based recognition technologies’ weaknesses, such as v ulnerability to illumination variations, large head poses, and spoofing at tacks. However, the accurate acquisition of the depth profile or surface i s an expensive and challenging process. We introduce a novel method to rec ognize faces from stereo camera systems without explicitly computing the f acial surface or depth map. Instead, the raw face stereo images along with the location in the image from which the face is extracted allow a convol utional neural network model (CNN) to improve the recognition task while a voiding the need to handle the geometric structure of the face explicitly. This way, we keep the simplicity and cost-efficiency of identity authenti cation from a single image while enjoying the benefits of geometric data w ithout explicitly reconstructing it. ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il LOCATION:Zoom Lecture: 94951674375 UID:123se24012024101800 END:VEVENT BEGIN:VEVENT DTSTART;TZID="Asia/Jerusalem":20220120T150000 DTEND;TZID="Asia/Jerusalem":20220120T160000 DTSTAMP;TZID="Asia/Jerusalem":20220120T150000 FREEBUSY;FBTYPE=BUSY:20220120T150000/20220120T160000 SUMMARY;LANGUAGE=en-US:phd talk by Stav Perle about Mathematical Technique s for Cryptanalysis at 2022-01-20 15:00:00 DESCRIPTION;LANGUAGE=en-US:Symmetric ciphers are cryptographic algorithms that use the same cryptographic keys for both encryption and decryption. T he key represents a shared secret between users, that is used to maintain a private information link. In our research we focus on cryptanalysis of b lock ciphers, which are the most widely used realization of symmetric ciph ers. Block ciphers are cryptosystems that consist of two algorithms, an encryption algorithm that accepts a symmetric key and a plaintext and outputs a ciphertext, and a decryption algorithm that reveals the plaintex t from the ciphertext and the key. Block ciphers are also the basis for ma ny cryptographic protocols. The security of these protocols relies on the security of the underlying block ciphers against attacks. Testing the security of block ciphers is highly important, both for stopping their us e if they are found insufficiently secure, and as design criteria how to d esign new more secure ciphers. Cryptanalysts use many techniques to examine the strength of block ciphers, ranging from differences and linear properties, through representation as polynomials, to more specialized st ructures, e.g., pairs, statistical characteristics and impossible events. The goal of the cryptanalyst is to design an attack on the block cipher wh ich is faster than exhaustive key search. In our research we combine some of these techniques with additional properties in order to improve the abi lity to examine the strength of block ciphers, and to design new better on es. ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il LOCATION:Zoom Lecture: 92017151302 UID:123se24012024101780 END:VEVENT BEGIN:VEVENT DTSTART;TZID="Asia/Jerusalem":20220124T100000 DTEND;TZID="Asia/Jerusalem":20220124T110000 DTSTAMP;TZID="Asia/Jerusalem":20220124T100000 FREEBUSY;FBTYPE=BUSY:20220124T100000/20220124T110000 SUMMARY;LANGUAGE=en-US:msc talk by Refael Cohen about SMEGA2: Distributed Deep Learning Using a Single Momentum Buffer at 2022-01-24 10:00:00 DESCRIPTION;LANGUAGE=en-US:As the field of deep learning progresses, and m odels become larger and larger, training deep neural networks has become a demanding task. The task requires a huge amount of compute power, and can still be very time consuming - especially when using just a single GPU. T o tackle this problem, distributed deep learning has come into play, with various asynchronous training algorithms. However, most of these algorithm s suffer from decreased accuracy as the number of workers increases. We in troduce a new method - Single MomEntum Gradient Accumulation ASGD (SMEGA2) , which outperforms existing methods in terms of final test accuracy and s cales up to as much as 64 asynchronous workers. ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il LOCATION:Zoom Lecture: 92984244781 UID:123se24012024101560 END:VEVENT BEGIN:VEVENT DTSTART;TZID="Asia/Jerusalem":20220124T150000 DTEND;TZID="Asia/Jerusalem":20220124T160000 DTSTAMP;TZID="Asia/Jerusalem":20220124T150000 FREEBUSY;FBTYPE=BUSY:20220124T150000/20220124T160000 SUMMARY;LANGUAGE=en-US:msc talk by Shadi Endrawis about Efficient Self-Sup ervised Data Collection for Offline Robot Learning at 2022-01-24 15:00:00 DESCRIPTION;LANGUAGE=en-US:a large batch of real or simulated robot intera ction data, using some data collection policy, and then learn from this da ta to perform various tasks, using offline learning algorithms. Previous w ork focused on manually designing the data collection policy, and on tasks where suitable policies can easily be designed, such as random picking po licies for collecting data about object grasping. For more complex tasks, however, it may be difficult to find a data collection policy that explore s the environment effectively, and produces data that is diverse enough fo r the downstream task. In this work, we propose that data collection policies should actively explore the environment to collect diverse data. In particular, we develop a simple-yet-effective goal-conditioned reinfor cement-learning method that actively focuses data collection on novel obse rvations, thereby collecting a diverse data-set. The method extends and im proves upon popular intrinsic motivation based methods for diverse explora tion. We evaluate our method on simulated robot manipulation tasks with vi sual inputs and show that our method leads to more diverse and evenly dist ributed data, and more importantly, that the data collection which activel y tries to reach novel states leads to significant improvements in the dow nstream learning tasks. ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il LOCATION:Zoom Lecture: 7446114621 UID:123se24012024101850 END:VEVENT BEGIN:VEVENT DTSTART;TZID="Asia/Jerusalem":20220124T150000 DTEND;TZID="Asia/Jerusalem":20220124T160000 DTSTAMP;TZID="Asia/Jerusalem":20220124T150000 FREEBUSY;FBTYPE=BUSY:20220124T150000/20220124T160000 SUMMARY;LANGUAGE=en-US:colloq talk by Oded Stein (MIT) about CS LECTURE: M athematical Foundations of Robust Geometry and Fabrication at 2022-01-24 1 5:00:00 DESCRIPTION;LANGUAGE=en-US:Current geometry methods for creating and ma nipulating shapes on computers can sometimes be unreliable and fail unpred ictably. Such failures make geometry tools hard to use, prevent non-expert s from creating geometry on their computers, and limit the use of geometry methods in domains where reliability is critical. We will discuss my rece nt efforts in proving when existing methods work as intended, my work in m aking methods more robust to imperfect input, my work in the creation of n ew reliable tools with mathematical guarantees, and my future efforts towa rds a reliable geometry pipeline. When used for computational fabrica tion, geometry methods can be expensive, finicky, and require a controlled environment. I will show how simple and economical manufacturing techniq ues can be used for computational fabrication by exploiting the geometric constraints inherent in specific materials and fabrication methods. We wil l take a look at how I create geometric tools to design for constrained fa brication techniques, and discuss how computational fabrication can be mad e both economical as well as accessible in difficult environments. Bi o: Oded Stein is a postdoc at MIT at the geometric data processing group . He obtained his MSc from ETH Zurich in 2015, and his PhD from Columbia U niversity in 2020. Oded is interested in geometry, computer graphics, and applied mathematics. He works on smoothness energies, partial differe ntial equations, discretization of geometric quantities, and their applica tions to computer graphics and digital fabrication.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il LOCATION:Zoom Lecture: 91550335554 UID:123se24012024101910 END:VEVENT BEGIN:VEVENT DTSTART;TZID="Asia/Jerusalem":20220125T103000 DTEND;TZID="Asia/Jerusalem":20220125T113000 DTSTAMP;TZID="Asia/Jerusalem":20220125T103000 FREEBUSY;FBTYPE=BUSY:20220125T103000/20220125T113000 SUMMARY;LANGUAGE=en-US:msc talk by Omer Rappoport about Solving Constraine d Horn Clauses Lazily and Incrementally at 2022-01-25 10:30:00 DESCRIPTION;LANGUAGE=en-US:Constrained Horn Clauses (CHCs) is a fragment o f First Order Logic (FOL), that has gained a lot of attention in recent ye ars. One of the main reasons for the rising interest in CHCs is the abilit y to reduce many verification problems to satisfiability of CHCs. For exam ple, program verification can naturally be described as the satisfiability of CHCs modulo a background theory such as linear arithmetic and arrays. To this end, CHC-solvers can be used as the back-end for different verific ation tools and allow to separate the generation of verification condition s from the decision procedure that decides if the verification conditions are correct or not. In our work, we present a novel framework, called LazyHorn, that is aimed at solving the satisfiability problem of CHCs mod ulo a background theory. The framework is driven by the idea that a set of CHCs can be solved in parts, making it an easier problem for the CHC-solv er. Furthermore, solving a set of CHCs can benefit from an interpretation revealed by the solver for its subsets. LazyHorn is lazy in that it gradua lly extends the set of checked CHCs, as needed. It is also incremental in its use of a CHC solver, supplying it with an interpretation, obtained for previously checked subsets. We have implemented an efficient instance of the framework that is restricted to a fragment of CHCs called linear CHCs. ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il LOCATION:Zoom Lecture: 93910185113 UID:123se24012024101860 END:VEVENT BEGIN:VEVENT DTSTART;TZID="Asia/Jerusalem":20220125T113000 DTEND;TZID="Asia/Jerusalem":20220125T123000 DTSTAMP;TZID="Asia/Jerusalem":20220125T113000 FREEBUSY;FBTYPE=BUSY:20220125T113000/20220125T123000 SUMMARY;LANGUAGE=en-US:pixel-club talk by Oren Nuriel (AWS) about Pixel Cl ub: TextAdaIN: Paying Attention to Shortcut Learning in TextRecognizers at 2022-01-25 11:30:00 DESCRIPTION;LANGUAGE=en-US:Leveragingthe characteristics of convolutional layers, neural networks are extremelyeffective for pattern recognition tas ks. However in some cases,their decisions are based on unintended informat ion leading to high performanceon standard benchmarks but also to a lack o f generalization to challengingtesting conditions and unintuitive failures . Recentworkhas termed this “shortcut learning” and addressed its presence in multipledomains. In text recognition, we reveal another such shortcut, whereby recognizersoverly depend on local image statistics. Motivated by this, we suggest anapproach to regulate the reliance on local statisticsth at improves text recognition performance. Ourmethod, termed TextAdaIN , creates local distortions in the feature map whichprevent the network fr om overfitting to localstatistics. It does so by viewing each feature map as a sequence of elementsand deliberately mismatching fine-grained feature statistics between elementsin a mini-batch. Despite TextAdaIN’s simplicit y, extensive experiments show its effectiveness compared to other, morecom plicated methods. TextAdaIN achieves state-of-the-art results on standardh andwritten text recognition benchmarks. Additionally, it generalizes tomul tiple architectures and to the domain of scene text recognition. Furthermo re, we demonstrate that integrating TextAdaINimproves robustness towards m ore challenging testing conditions. Short bio: Oren Nurielis an app lied computer vision scientist at AWS. He holds an MSc degree inComputer S cience from the Tel-Aviv University. ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il LOCATION:Zoom Lecture: https://technion.zoom.us/my/chaimbaskin UID:123se24012024101920 END:VEVENT BEGIN:VEVENT DTSTART;TZID="Asia/Jerusalem":20220126T103000 DTEND;TZID="Asia/Jerusalem":20220126T113000 DTSTAMP;TZID="Asia/Jerusalem":20220126T103000 FREEBUSY;FBTYPE=BUSY:20220126T103000/20220126T113000 SUMMARY;LANGUAGE=en-US:msc talk by Gal Shalom about pISTA: preconditioned Iterative Soft Thresholding Algorithm for Graphical Lasso at 2022-01-26 10 :30:00 DESCRIPTION;LANGUAGE=en-US:We propose a novel quasi-Newton method for s olving the sparse inverse covariance estimation problem also known as the graphical least absolute shrinkage and selection operator (GLASSO). T his problem is often solved using a second order quadratic approximation. However, in such algorithms the Hessian term is complex and computationall y expensive to handle. To this end,our algorithm uses the inverse of the H essian as a preconditioner to simplify and approximate the quadratic eleme nt at the cost of a more complex l1 element. The variables of the resultin g preconditioned problem are coupled only by the l1 sub-derivative of each other, which can be guessed with minimal cost using the gradient itself, allowing the algorithm to be parallelized and implemented efficiently on G PU hardware accelerators. Numerical results on synthetic and real data dem onstrate that our method is competitive with other state-of-the-art approa ches.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il LOCATION:Zoom Lecture: 91228689582 UID:123se24012024101700 END:VEVENT BEGIN:VEVENT DTSTART;TZID="Asia/Jerusalem":20220126T113000 DTEND;TZID="Asia/Jerusalem":20220126T123000 DTSTAMP;TZID="Asia/Jerusalem":20220126T113000 FREEBUSY;FBTYPE=BUSY:20220126T113000/20220126T123000 SUMMARY;LANGUAGE=en-US:ceClub talk by Haggai Eran (EE, Technion) about ceC lub: SmartNIC Inline Processing at 2022-01-26 11:30:00 DESCRIPTION;LANGUAGE=en-US:The inline processing technique enables data tr ansformation as a system transfers data to or from a processing node. It i s used to offload computations and accelerate data-intensive communication tasks, reducing latency and power due to data movement and improving thro ughput by using the best processing core for the job. However, inline proc essing poses several challenges: it breaks existing operating system and n etwork stack layers and makes it difficult to reuse previous software and hardware. This talk presents new operating system abstractions to allow ap plication layer inline processing on SmartNICs and new SmartNIC designs, e nabling fine-grain hardware virtualization and reusing existing ASIC NIC f unctionality with FPGA-based SmartNICs. In addition, I will present a comp lementary line of research that uses SmartNICs to accelerate data-center n etworks transparently and enable sharing of competing network transports. *PhD student under supervision of Prof. Mark Silberstein. ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il LOCATION:Zoom Lecture: 97164378341 UID:123se24012024101930 END:VEVENT BEGIN:VEVENT DTSTART;TZID="Asia/Jerusalem":20220130T133000 DTEND;TZID="Asia/Jerusalem":20220130T143000 DTSTAMP;TZID="Asia/Jerusalem":20220130T133000 FREEBUSY;FBTYPE=BUSY:20220130T133000/20220130T143000 SUMMARY;LANGUAGE=en-US:cggc talk by Alon Rashelbach (EE, Technion) about C GGC Seminar: Trading Memory for Computations: Scaling Range Matching on th e Critical Path at 2022-01-30 13:30:00 DESCRIPTION;LANGUAGE=en-US:Range matching (RM) is a crucial component in c omputer systems, widely used in address translation, hard drives, network switches, and many more applications. RM is performed whenever one wishes to locate a range that contains an input number, given a large set of rang es. Any page-based mechanism uses RM, as pages are basically ranges. Longe st prefix matching (LPM) uses ternary rules, which are also ranges. Firewa lls are one example of multidimensional RM since ACL rules consist of seve ral wildcarded fields that can be represented as ranges. Existing algorith ms for RM are either page-based, tree-based, or hash-table-based. Either w ay, they all rely on pointer-chasing techniques, which impede their scalab ility to large range sets that spill out of the CPU core cache. Furthermor e, their inherent lack of access locality and the data-dependent nature of their memory accesses make hardware prefetchers ineffective. Our res earch focuses on a new data structure, the Range Query Recursive Model Ind ex (RQ-RMI). RQ-RMI uses shallow neural-nets (NN) for learning the distrib ution of a given set of ranges, turning the costly lookup operations into NN inference. Crucially, the RQ-RMI training algorithm guarantees a tight bound on its lookup latency, ensures its correctness, and converges fast: it can index 500K ranges in a few hundred milliseconds on a single CPU cor e. We develop NuevoMatch (NM), an algorithm for multi-field packet cl assification, and integrate it into the critical path of the popular, open -source, virtual switch Open vSwitch (OVS). NM scales OVS to support 500K OpenFlow rules with a 12.3X geomean throughput speedup and up to 50K updat es per second. ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il LOCATION:Zoom Lecture: 96379418284 UID:123se24012024101530 END:VEVENT BEGIN:VEVENT DTSTART;TZID="Asia/Jerusalem":20220201T133000 DTEND;TZID="Asia/Jerusalem":20220201T143000 DTSTAMP;TZID="Asia/Jerusalem":20220201T133000 FREEBUSY;FBTYPE=BUSY:20220201T133000/20220201T143000 SUMMARY;LANGUAGE=en-US:msc talk by Roei Kisous about Clustering Based Data Migration in Deduplicated Storage at 2022-02-01 13:30:00 DESCRIPTION;LANGUAGE=en-US:Deduplication is a leading method for reducing physical storage capacity when duplicate data is present. This method can be applied on chunks, files, containers, and more. Instead of storing the same physical data multiple times, a pointer is created from each logical copy to the same physical copy, saving the space of the duplicate data. Du e to this, data is shared between objects, such as files or entire directo ries, which result in garbage collection overhead and migration challenges . In our work, we addressed the general migration problem where files are remapped between different volumes due to system expansion or mainten ance. The question of which files and blocks to migrate has been extensiv ely studied in systems without deduplication. However, only simplified mig ration problems have been considered in the context of deduplicated storag e. As part of a migration plan, we aim to minimize the system's size whil e simultaneously ensuring that the storage load is evenly distributed acro ss the volumes and that the network traffic required for the migration doe s not exceed its allocation. Following that, we outline a way to deve lop effective migration plans using hierarchical clustering. Clustering re fers to grouping objects based on their similarity. Hierarchical clusterin g, in particular, takes the distance between those objects into account. E ach object is initially clustered separately, and the process of iterative clustering merges, in each step, two clusters with a minimal distance bet ween them. We are interested in clustering files with high similarity toge ther in order to reduce the amount of physical data while still maintainin g low network traffic and a balanced system. Based on each cluster, we cal culate data savings, traffic consumed, and load balance achieved and deter mine the plan's quality. We show that this method has different trade offs between computation time and migration efficiency compared to other a lgorithms such as greedy and ILP (Integer Linear Programming). Our algorit hm achieves almost identical results (and sometimes even better) than ILP, which, theoretically, is optimal, but in much less time. ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il LOCATION:Zoom Lecture: 8183278482 UID:123se24012024101660 END:VEVENT BEGIN:VEVENT DTSTART;TZID="Asia/Jerusalem":20220206T120000 DTEND;TZID="Asia/Jerusalem":20220206T130000 DTSTAMP;TZID="Asia/Jerusalem":20220206T120000 FREEBUSY;FBTYPE=BUSY:20220206T120000/20220206T130000 SUMMARY;LANGUAGE=en-US:msc talk by Ariel Kolikant about ILP Based Load Bal ancing in Deduplicated Storage Systems at 2022-02-06 12:00:00 DESCRIPTION;LANGUAGE=en-US:Deduplication reduces the size of the data s tored in large-scale storage systems by replacing duplicate data blocks wi th references to their unique copies. This creates dependencies between fi les that contain similar content and complicates the management of data in the system. In the work presented in this seminar, we have addressed the problem of data migration, where files are remapped between different volu mes because of system expansion or maintenance. The challenge of determini ng which files and blocks to migrate has been studied extensively for syst ems without deduplication. In the context of deduplicated storage, however , only simplified migration scenarios were considered. In our work we have formulated the general migration problem for deduplicated systems as an o ptimization problem whose objective is to minimize the system’s size while ensuring that the storage load is evenly distributed between the system’s volumes, and that the network traffic required for the migration does not exceed its allocation. We then modeled an ILP algorithm to solve the migr ation problem generated, and compared it’s results to two other algorithms solving the same generated migration problem: the greedy algorithm and th e clustering algorithm. Our ILP algorithm manages to consistently obtain t he best solutions to the problem though it requires significantly larger e xecution times.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il LOCATION:Zoom Lecture: 97413372304 UID:123se24012024101900 END:VEVENT BEGIN:VEVENT DTSTART;TZID="Asia/Jerusalem":20220208T103000 DTEND;TZID="Asia/Jerusalem":20220208T113000 DTSTAMP;TZID="Asia/Jerusalem":20220208T103000 FREEBUSY;FBTYPE=BUSY:20220208T103000/20220208T113000 SUMMARY;LANGUAGE=en-US:pixel-club talk by Aviad Aberdam (CS/EE, Technion a bout Pixel Club: Convex Optimization: Adaptive Learned Solvers and Coordin ateGradient Descent at 2022-02-08 10:30:00 DESCRIPTION;LANGUAGE=en-US:Neural networks that are based on unfolding of an iterativesolver, such as LISTA (learned iterative soft threshold algori thm), are widelyused due to their accelerated performance. Nevertheless, a s opposed tonon-learned solvers, these networks are trained on a certain d ictionary, andtherefore they are inapplicable for varying model scenarios. This talkintroduces an adaptive learned solver, termed Ada-LISTA, which r eceives pairsof signals and their corresponding dictionaries as inputs, an d learns auniversal architecture to serve them all. We prove that this sch eme isguaranteed to solve sparse coding in linear rate for varying models, includingdictionary perturbations and permutations. We also provide an ex tensivenumerical study demonstrating its practical adaptation capabilities . Anotherway to improve performance is to adopt coordinate descent algorit hms which arepopular in machine learning and large-scale data analysis pro blems due to theirlow computational cost. In this talk, we define a monoto ne acceleratedc ordinate gradient descent-type method for problems consist ing of minimizing f+ g, where f is quadratic and g is nonsmooth and non-se parable. The algorithmis enabled by employing the forward–backward envelop e, a composite envelopethat possess an exact smooth reformulation of f + g . We prove the algorithmachieves a convergence rate of O(1/k^1.5) in terms of the original objective function,improving current coordinate descent-t ype algorithms. In addition, we describean adaptive variant of the algorit hm that backtracks the spectral informationand coordinate Lipschitz consta nts of the problem. We numerically examine ouralgorithms on various settin gs, including two-dimensional total-variation-basedimage inpainting proble ms, showing a clear advantage in performance overcurrent coordinate descen t-type methods. PHD student under supervision of Prof. Michael Elad . ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il LOCATION:Zoom Lecture: https://technion.zoom.us/my/chaimbaskin UID:123se24012024101950 END:VEVENT BEGIN:VEVENT DTSTART;TZID="Asia/Jerusalem":20220208T113000 DTEND;TZID="Asia/Jerusalem":20220208T123000 DTSTAMP;TZID="Asia/Jerusalem":20220208T113000 FREEBUSY;FBTYPE=BUSY:20220208T113000/20220208T123000 SUMMARY;LANGUAGE=en-US:msc talk by Adi Mesika about CloudWalker: 3D Point Cloud Learning by Random Walks for Shape Analysis at 2022-02-08 11:30:00 DESCRIPTION;LANGUAGE=en-US:Point clouds are gaining prominence as a method for representing 3D shapes, but their irregular structure poses a challen ge for deep learning methods. In this paper we propose CloudWalker, a novel method for learning 3D shapes using random walks. Previous works a ttempt to adapt Convolutional Neural Networks (CNNS) or impose a grid or m esh structure to 3D point clouds. This work presents a different approach for representing and learning the shape from a given point set. The key id ea is to impose structure on the point set by multiple random walks throug h the cloud for exploring different regions of the 3D object. Then we lear n a per-point and per-walk representation and aggregate multiple walk pred ictions at inference. Our approach achieves state-of-the-art results for two 3D shape analysis tasks: classification and retrieval. Furthermo re, we propose a shape ambiguity indicator function that uses cross-walk a nd inter-walk variance measures to subdivid the shape space. ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il LOCATION:Zoom Lecture: https://technion.zoom.us/my/chaimbaskin UID:123se24012024101940 END:VEVENT BEGIN:VEVENT DTSTART;TZID="Asia/Jerusalem":20220221T093000 DTEND;TZID="Asia/Jerusalem":20220221T103000 DTSTAMP;TZID="Asia/Jerusalem":20220221T093000 FREEBUSY;FBTYPE=BUSY:20220221T093000/20220221T103000 SUMMARY;LANGUAGE=en-US:msc talk by Gal Sidi about DELETE: Using deep learn ing to minimize latency in CEP systems at 2022-02-21 09:30:00 DESCRIPTION;LANGUAGE=en-US:The ability to detect complex patterns in massi ve data streams is critical for many real-time applications. These applica tions must uphold low latency requirements, delivering alerts and notifica tions with minimal response delays. Complex event processing (CEP), a lead ing technology for performing this task, is suitable for the efficient and robust detection of complex patterns. However, the CEP complexity grows e xponentially with respect to the length of the pattern and the intensity o f the data stream. As a result, most CEP based applications that monitor f requent data are often incapable of analyzing complex patterns under real- time conditions, and thus define only simple patterns in practice. We pres ent a novel deep learning based framework to accelerate the detection proc ess of CEP systems. We use a neural network to filter the input stream, th ereby significantly reducing the detection latency. This network is traine d to efficiently filter the input based on the pattern. In parallel to run ning a CEP engine that processes the filtered substream, we employ CEP on the original input, which ensures a detection accuracy of 100%. Extensive experimental evaluation of several real-world datasets shows that our appr oach consistently improves the latency by up to a factor of 100 as compare d to existing state-of-the-art CEP systems. ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il LOCATION:Zoom Lecture: 98932630687 UID:123se24012024101970 END:VEVENT BEGIN:VEVENT DTSTART;TZID="Asia/Jerusalem":20220223T090000 DTEND;TZID="Asia/Jerusalem":20220223T100000 DTSTAMP;TZID="Asia/Jerusalem":20220223T090000 FREEBUSY;FBTYPE=BUSY:20220223T090000/20220223T100000 SUMMARY;LANGUAGE=en-US:msc talk by Guy Shapira about REDEEMER: Reinforceme nt Learning Based CEP Pattern Miner for Knowledge Extraction at 2022-02-23 09:00:00 DESCRIPTION;LANGUAGE=en-US:Complex Event Processing (CEP) are a set of met hods that allow efficient knowledge extraction from massive data streams u sing complex and highly descriptive patterns. As of today, in many fields, patterns are manually defined by human experts. However, desired patterns often contain convoluted relations that are difficult for humans to detec t, and human expertise is scarce in many domains. We present REDEEMER , a novel reinforcement and active learning approach aimed at mining CEP p atterns that allow expansion of the knowledge extracted while reducing the human effort required. This approach includes a novel policy gradien t method for vast multivariate spaces and a new way to combine reinforceme nt and active learning for CEP rule learning while minimizing the amount of labels needed for training. Our experiments on diverse data-set s demonstrate that REDEEMER is able to extend pattern knowledge while outp erforming several state-of-the-art reinforcement learning methods for patt ern mining. ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il LOCATION:Zoom Lecture: 94383091556 UID:123se24012024101980 END:VEVENT BEGIN:VEVENT DTSTART;TZID="Asia/Jerusalem":20220224T083000 DTEND;TZID="Asia/Jerusalem":20220224T173000 DTSTAMP;TZID="Asia/Jerusalem":20220224T083000 FREEBUSY;FBTYPE=BUSY:20220224T083000/20220224T173000 SUMMARY;LANGUAGE=en-US:CSpecial Event about TCE-MLIS 2021 Conference at 2 022-02-24 08:30:00 DESCRIPTION;LANGUAGE=en-US:MLIS, the Technion AI center, in collaboration with TCE, would like to invite you to participate in the annual MILS-TCE conference. AI is now a major buzz word everywhere and expectations are sky-rocketing, b ut what is true state-of-the-art and what can be actually implemented in t he AI and Machine Learning fields? In a series of lectures, Technion researchers will present cutting-edge AI research conducted in the instit ution. Simultaneously, in thematic workshops we will explore the role of A I and data-science in the Educational and Urban Planning fields. More details, registration and program. Do not miss the EARLY BIRD registration by February 1, 2022. ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il LOCATION:ELMA Arts Complex, Zichron Ya'acov UID:123se24012024101750 END:VEVENT BEGIN:VEVENT DTSTART;TZID="Asia/Jerusalem":20220224T100000 DTEND;TZID="Asia/Jerusalem":20220224T160000 DTSTAMP;TZID="Asia/Jerusalem":20220224T100000 FREEBUSY;FBTYPE=BUSY:20220224T100000/20220224T160000 SUMMARY;LANGUAGE=en-US:CSpecial Event about A Meeting with Potential Stude nts for Technion CS Studies at 2022-02-24 10:00:00 DESCRIPTION;LANGUAGE=en-US:A meeting with potential stu dents who are interested in studies at the Technion and the Faculty of Computer Science will be held online on CS Facebook, onThursday, February 24, 2022, at 10:100 and at 14:00. Details and registration. ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il LOCATION:Zoom Event: Registration UID:123se24012024102010 END:VEVENT BEGIN:VEVENT DTSTART;TZID="Asia/Jerusalem":20220228T113000 DTEND;TZID="Asia/Jerusalem":20220228T123000 DTSTAMP;TZID="Asia/Jerusalem":20220228T113000 FREEBUSY;FBTYPE=BUSY:20220228T113000/20220228T123000 SUMMARY;LANGUAGE=en-US:msc talk by Antonio Abu Nassar about Semantic Symme try in Transducers at 2022-02-28 11:30:00 DESCRIPTION;LANGUAGE=en-US:In model checking, we work toward deciding whet her a system satisfies a given specification. Often, a system exhibits som e type of symmetry in its structure or in its behaviour. Such symmetries c an be exploited by a designer to alleviate some of the complexity of model checking, as well as to gain insight into the behaviour of the system. Th us, we want to decide whether a given system exhibits symmetry. Symme try is not a well-defined concept and might come in various forms, each ca pturing a different characteristic behaviour. In this talk, I introduce se veral notions of semantic symmetry in transducers, and demonstrate the def initions and the behaviours they capture, as well as pose the algorithmic questions pertaining to them and their solutions. In particular, I pr esent the notion of simulation by rounds, whose usefulness is in that it c an be applied to process symmetry. In this setting, words are partitioned into rounds, and a transducer is round simulated by another if for every i nput word, we can shuffle the letters within each round such that the outp ut of the simulating transducer on the shuffled word is itself a shuffle o f the output of the simulated transducer. We use tools and techniques from logic, algebra and automata theory. ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il LOCATION:Zoom Lecture: 98340908086 UID:123se24012024101960 END:VEVENT BEGIN:VEVENT DTSTART;TZID="Asia/Jerusalem":20220301T110000 DTEND;TZID="Asia/Jerusalem":20220301T120000 DTSTAMP;TZID="Asia/Jerusalem":20220301T110000 FREEBUSY;FBTYPE=BUSY:20220301T110000/20220301T120000 SUMMARY;LANGUAGE=en-US:msc talk by Or Feldman about Improving Graph Neural Networks Expressivity Via Spectral and Combinatorial Pre-Colorings at 202 2-03-01 11:00:00 DESCRIPTION;LANGUAGE=en-US:Graph isomorphism testing is usually approached via the comparison of graph invariants. Two popular alternatives that off er a good trade-off between expressive power and computational efficiency are combinatorial (i.e., obtained via the Weisfeiler-Leman (WL) test) and spectral invariants. While the exact power of the latter is still an open question, the former is regularly criticized for its limited power, when a standard configuration of uniform pre-coloring is used. This drawback hin ders the applicability of Message Passing Graph Neural Networks (MPGNNs), whose expressive power is upper bounded by the WL test. Relaxing the assum ption of uniform pre-coloring, we show that one can increase the expressiv e power of the WL test ad infinitum. Following that, we propose an efficie nt pre-coloring based on spectral features that provably increases the exp ressive power of the vanilla WL test. ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il LOCATION:Zoom Lecture: 96864188946 UID:123se24012024101990 END:VEVENT BEGIN:VEVENT DTSTART;TZID="Asia/Jerusalem":20220302T113000 DTEND;TZID="Asia/Jerusalem":20220302T123000 DTSTAMP;TZID="Asia/Jerusalem":20220302T113000 FREEBUSY;FBTYPE=BUSY:20220302T113000/20220302T123000 SUMMARY;LANGUAGE=en-US:ceClub talk by Farhad Merchant (Aachen University) about ceClub: Embedded Systems and Security at 2022-03-02 11:30:00 DESCRIPTION;LANGUAGE=en-US:Protecting intellectual properties from untrust ed design houses and foundries has become highly challenging. In this talk , I will focus on the security aspects of embedded systems. First, I will discuss the logic locking tools and attack methods developed at the Instit ute for Communication Technologies and Embedded Systems, RWTH Aachen Unive rsity. In the second part of the talk, I will focus on developing low-powe r, high-performance embedded systems based on emerging non-volatile memory technologies. These methods incorporate the non-Von Neumann computing mod el, where the computations are performed inside memory to reduce the data movements. Finally, I will touch upon the security aspects of emerging tec hnologies. Bio: Farhad Merchant received his Ph.D. from the Indian Institute of Science, Bangalore (India), in 2016. His Ph.D. thesis title was "Algorithm-Architecture Co-design for Dense Linear Algebra Computation s." He received the DAAD fellowship during his PhD. He worked as a postdoc toral research fellow at Nanyang Technological University (NTU), Singapore , from March 2016 to December 2016. In December 2016, he moved to Corporat e Research in Robert Bosch in Bangalore as a Researcher, where he worked o n numerical methods for ordinary differential equations. He joined Institu te for Communication Technologies and Embedded Systems, RWTH Aachen Univer sity, in December 2017 as a postdoctoral research fellow in the Chair for Software for Systems on Silicon. Farhad is the recipient of the HiPEAC tec hnology transfer award in 2019. His research interests are hardware securi ty and neuromorphic computing. ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il LOCATION:Room 861, EE Meyer Building and zoom Lecture: 96393404383 UID:123se24012024102000 END:VEVENT BEGIN:VEVENT DTSTART;TZID="Asia/Jerusalem":20220309T110000 DTEND;TZID="Asia/Jerusalem":20220309T120000 DTSTAMP;TZID="Asia/Jerusalem":20220309T110000 FREEBUSY;FBTYPE=BUSY:20220309T110000/20220309T120000 SUMMARY;LANGUAGE=en-US:msc talk by Omer Antverg about Analyzing Individual Neurons in Language Models at 2022-03-09 11:00:00 DESCRIPTION;LANGUAGE=en-US:Neural language models have significantly devel oped in recent years, becoming more and more successful on numerous langua ge tasks. Those models rely on encoding words as hidden vector representat ions, before utilizing these representations for the task at hand. Their s uccess spiked interest in their interpretability: understanding how they w ork, and what is encoded within these representations. While many studies have shown that linguistic information is encoded in hidden word represent ations, few have studied individual neurons of these representations, to s how how and in which neurons it is encoded. Among these studies, the commo n approach is to use an external probe to rank neurons according to their relevance to some linguistic attribute, and to evaluate the obtained ranki ng using the same probe that produced it. We show two pitfalls in this met hodology: 1. It confounds distinct factors: probe quality and ranking qual ity. We follow this methodology and show where the conflation occurs, and why it may mislead us. 2. It focuses on encoded information, rather than i nformation that is used by the model. We perform neuron-level causal analy sis of the model, and show that encoded information and used information a re not the same. We compare two recent ranking methods and a simple one we introduce, and evaluate them with regard to both of these aspects. ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il LOCATION:Zoom Lecture: 93322539388 UID:123se24012024102020 END:VEVENT BEGIN:VEVENT DTSTART;TZID="Asia/Jerusalem":20220310T143000 DTEND;TZID="Asia/Jerusalem":20220310T153000 DTSTAMP;TZID="Asia/Jerusalem":20220310T143000 FREEBUSY;FBTYPE=BUSY:20220310T143000/20220310T153000 SUMMARY;LANGUAGE=en-US:msc talk by Idan Raz about Model-Based Simulation f or SMT Cores at 2022-03-10 14:30:00 DESCRIPTION;LANGUAGE=en-US:Studies that evaluate new architectural designs of virtual memory typically employ a ``model-based’’ methodology that rel ies on simulations of the translation lookaside buffer (TLB) coupled with empirical performance models. We observe that this methodology is limited in that each simulated thread of execution has its own dedicated TLB, wher eas modern processors share a single TLB among multiple threads through `` simultaneous multithreading’’ (SMT). Existing model-based research is thus unable to explore virtual memory designs in SMT context. We address this problem: (1) by showing that the behavior of different multiprogrammed thr ead combinations varies over time nontrivially, and by introducing a syste matic approach for measuring this behavior with bounded error; (2) by deve loping a TLB simulator capable of realistically combining multiple memory- reference streams (of the SMT threads) into one; (3) by validating the sim ulator’s accuracy against real (Intel) processors to ensure the correctnes s of our approach, which required us to reverse engineer their TLB evictio n policy; and (4) by showing how to build empirical models that predict ru ntimes of different SMT combinations from their combined simulated TLB mis s rate. We demonstrate our methodology’s usefulness by evaluating a new TL B partitioning mechanism for SMT processor cores. ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il LOCATION:Zoom Lecture: 91575675908 UID:123se24012024102050 END:VEVENT BEGIN:VEVENT DTSTART;TZID="Asia/Jerusalem":20220310T153000 DTEND;TZID="Asia/Jerusalem":20220310T163000 DTSTAMP;TZID="Asia/Jerusalem":20220310T153000 FREEBUSY;FBTYPE=BUSY:20220310T153000/20220310T163000 SUMMARY;LANGUAGE=en-US:msc talk by Maxim Barsky about PartTLB: Dynamic TLB Partitioning for SMT Systems at 2022-03-10 15:30:00 DESCRIPTION;LANGUAGE=en-US:Simultaneous multithreading (SMT) increases the cost of memory address translation due to sharing of the translation look aside buffer (TLB) among multiple threads. Current x86 processors use a `` competitively-shared’’ TLB, in which low-locality threads might needlessly waste TLB resources and thus degrade the performance of neighboring high- locality threads. To address this problem, we introduce PartTLB, a new mec hanism that: (1) samples the TLB requests of the competing threads, (2) pr edicts their miss rate with different TLB sizes, and (3) partitions the TL B accordingly to minimize the overall miss rate. PartTLB requires small ad ditional hardware (about 1% of the TLB size) and no software involvement. For a 2-way SMT, PartTLB improves performance by up to 37% for one thread and worsens performance by up to 3% for the other thread. ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il LOCATION:Zoom Lecture: 93840198625 UID:123se24012024102040 END:VEVENT BEGIN:VEVENT DTSTART;TZID="Asia/Jerusalem":20220315T110000 DTEND;TZID="Asia/Jerusalem":20220315T140000 DTSTAMP;TZID="Asia/Jerusalem":20220315T110000 FREEBUSY;FBTYPE=BUSY:20220315T110000/20220315T140000 SUMMARY;LANGUAGE=en-US:CSpecial Event about CS Orientation Day 2022 at 202 2-03-15 11:00:00 DESCRIPTION;LANGUAGE=en-US:CS 2022 Orientation Day for new students will b e held on Tuesday, March 15, 2022, and will begin at 10:00 with a Technion meeting at the Churchill Auditorium where the Senior Vice President, the Dean for Undergraduate Studies and the Students Dean and Chairman of the T echnion Student Association will speak to the new students, and between 11 :00-14:00 there will be a gathering at the Computer Science Taub Building, and a meeting at the Taub 1 Auditorium in the entrance floor, which will include tours, receiving a student card, acquaintance with members of the academic staff and the undergrads secretariat, as well as a meeting with s tudent association representatives, faculty tours, demonstrations and more .The prize-winni
ng competition will last about 30 hours, registration is open and the num
ber of places is limited.
More details and registration.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:CS Taub Building
UID:123se24012024102160
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20220524T170000
DTEND;TZID="Asia/Jerusalem":20220524T190000
DTSTAMP;TZID="Asia/Jerusalem":20220524T170000
FREEBUSY;FBTYPE=BUSY:20220524T170000/20220524T190000
SUMMARY;LANGUAGE=en-US:CSpecial Event about Microsoft Lecture: Improving P
roductivity Through NLP at 2022-05-24 17:00:00
DESCRIPTION;LANGUAGE=en-US:You are invited to a technological lecture by D
ikla Dotan Cohen, Director of Research at Microsoft, on Improving Producti
vity Through NLP, which will present the methodology of Office 365 product
s and the challenges it poses, on Tuesday, May 24, 17:00, in Room 337, CS
Taub Building.
Please pre-register.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Room 337 Taub Bld.
UID:123se24012024102410
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20220531T113000
DTEND;TZID="Asia/Jerusalem":20220531T123000
DTSTAMP;TZID="Asia/Jerusalem":20220531T113000
FREEBUSY;FBTYPE=BUSY:20220531T113000/20220531T123000
SUMMARY;LANGUAGE=en-US:pixel-club talk by Ron Mokady (Tel-Aviv University)
about Pixel Club: Moving Forward with StyleGAN to Real Data and New Domai
ns at 2022-05-31 11:30:00
DESCRIPTION;LANGUAGE=en-US:StyleGAN is already quite famous for its unrema
rkable image editing capabilities. Although other generative models (e.g.
diffusion models) achieve comparable synthesis quality, they cannot reprod
uce these semantically richmanipulations. In particular, StyleGAN allows t
he modification of various attributes, such as hair, age, pose, expression
, and make-up, while still maintaining a high level of realism.
Yet,
it is still challenging to leverage these traits for real data or new doma
ins. In this talk, we discuss three main obstacles: First, how to edit rea
l images using latent-based manipulation, i.e., inverting a real image to
StyleGAN'slatent space. Second, how to employ the per-image editing over v
ideos, which requires another dimension of realism - temporal consistency.
Lastly, we explore the possibilities of training StyleGAN in new and exci
ting domains.
Short bio:
I am a Computer Science Ph.D. student at
Tel-Aviv University, under the supervision of Prof. Daniel Cohen-Or and D
r. Amit H. Bermano. Previously, I spent the 2020 summer at FAIR under the
supervision of Prof. Lior Wolf, and the 2021 summer at Google Research. My
main research interest is machine learning applications for computer visi
on and graphics. In particular, I work on image and video synthesis, while
also interested in disentanglement, temporal coherence, supervision reduc
tion, and the utilization of pre-trained models.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture: https://technion.zoom.us/my/chaimbaskin
UID:123se24012024102430
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20220601T103000
DTEND;TZID="Asia/Jerusalem":20220601T143000
DTSTAMP;TZID="Asia/Jerusalem":20220601T103000
FREEBUSY;FBTYPE=BUSY:20220601T103000/20220601T143000
SUMMARY;LANGUAGE=en-US:CSpecial Event about Recruitment Day by Vayyar at 2
022-06-01 10:30:00
DESCRIPTION;LANGUAGE=en-US:Vayyar representatives will visit CS to demonst
rate 3D sensor technologies capable of seeing through objects, and for a l
ecture on 13:30 by Uri Adar, Vayyar system group member, about 3D imaging
in RF, on Wednesday, June 1, 2022, between 10:30-14:30, at the CS Taub Lob
by. More details in the attached poster.
Please pre-register for the lecture.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:
UID:123se24012024102500
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20220601T123000
DTEND;TZID="Asia/Jerusalem":20220601T133000
DTSTAMP;TZID="Asia/Jerusalem":20220601T123000
FREEBUSY;FBTYPE=BUSY:20220601T123000/20220601T133000
SUMMARY;LANGUAGE=en-US:Theory Seminar talk by Natan Rubin (Ben-Gurion Univ
ersity) about Theory Seminar: From Selection Theorems to Weak Epsilon-Nets
in Higher Dimensions (and back?) at 2022-06-01 12:30:00
DESCRIPTION;LANGUAGE=en-US:Given a finite point set $P$ in $R^d$, and $\ep
s>0$ we say that a point set $N$ in $R^d$ is a weak $\eps$-net if it pierc
es every convex set $K$ with $|K\cap P|\geq \eps |P|$.
Let $d\geq 3$. We
show that for any finite point set in $R^d$, and any $\eps>0$, there exis
ts a weak $\eps$-net of cardinality $o(1/\eps^{d-1/2})$. Here $delta>0$ is
an arbitrary small constant.
This is the first improvement of the bound
of $O^*(1/\eps^d)$ that was obtained in 1993 by Chazelle, Edelsbrunner, G
rigni, Guibas, Sharir, and Welzl for general point sets in dimension $d\ge
q 3$. The study of weak epsilon-nets is closely related to the fundamental
problem of finding a point that hits many simplices in a dense $(d+1)$-un
iform geometric hypergraph.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 201
UID:123se24012024102490
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20220607T113000
DTEND;TZID="Asia/Jerusalem":20220607T123000
DTSTAMP;TZID="Asia/Jerusalem":20220607T113000
FREEBUSY;FBTYPE=BUSY:20220607T113000/20220607T123000
SUMMARY;LANGUAGE=en-US:msc talk by Ben Finkelshtein about Robustness and R
otation Equivariance in Geometric Deep Learning at 2022-06-07 11:30:00
DESCRIPTION;LANGUAGE=en-US:Graph neural networks (GNNs) have shown broad a
pplicability in a variety of domains.
These domains, e.g., social net
works and recommendation systems, are fertile ground for malicious users a
nd behavior. In a series of works, we study the robustness of GNNs under d
ifferent scenarios and present a simple rotation and permutation equivaria
nt point-cloud GNN.
We show that GNNs are vulnerable to the scenario
of strategic behavior of multiple users (i.e., Strategic Classification)
and to the extremely limited (and thus quite realistic) scenario of a sing
le-node adversarial attack. That is, an attacker can force the GNN to clas
sify any target node to a chosen label, by only slightly perturbing the fe
atures or the neighbors' list of another single arbitrary node in the grap
h, even when not being able to select that specific attacker node.
A
dditionally, as equivariance to permutations and rigid motions is an impor
tant inductive bias for various 3D learning problems -- we present a simpl
e rotation and permutation equivariant point-cloud network that can approx
imate any equivariant function.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture: https://technion.zoom.us/my/benfin
UID:123se24012024102460
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20220607T183000
DTEND;TZID="Asia/Jerusalem":20220607T203000
DTSTAMP;TZID="Asia/Jerusalem":20220607T183000
FREEBUSY;FBTYPE=BUSY:20220607T183000/20220607T203000
SUMMARY;LANGUAGE=en-US:CSpecial Event about Online Meeting: How to be "On
it" Financially at 2022-06-07 18:30:00
DESCRIPTION;LANGUAGE=en-US:You are invited to an online meeting led by Yae
l Marom, Savings Product Manager at RiseUp, on: How to be "on it" financia
lly, which will explain in depth what our monthly financial situation is a
nd the way to know how much we really need to spend, how to prepare ahead,
for short term and long term, in order to gradually build our economic gr
owth, on Tuesday, June 7, at 18:30.
For link to the meeting pre-register.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Event: Registration
UID:123se24012024102350
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20220608T100000
DTEND;TZID="Asia/Jerusalem":20220608T110000
DTSTAMP;TZID="Asia/Jerusalem":20220608T100000
FREEBUSY;FBTYPE=BUSY:20220608T100000/20220608T110000
SUMMARY;LANGUAGE=en-US:cggc talk by Shir Rorberg (CS, Technion) about CGGC
Seminar: Nozzle Modification for Efficient FDM 3D Printing at 2022-06-08
10:00:00
DESCRIPTION;LANGUAGE=en-US:3D printing is based on layered manufacturing,
where the layers are printed consecutively in increasing height order. In
Fused Depositing Modeling (FDM), the printing head may travel without extr
uding material between separated “islands” of the sliced layers. These tra
vel movements increase the printing time and reduce the quality of the 3D
printed part. We present an extended nozzle modification, which can be app
lied to off-the-shelf FDM printers, and a corresponding toolpath generatio
n algorithm. Together, these dramatically reduce the amount of travel move
ment, thus improving the printing time and quality of the results. The ext
ended nozzle allows us to print the sliced layers out of order, where part
of a lower layer might be printed after a higher layer was already printe
d. Our toolpath generation algorithm takes advantage of this capability an
d generates a toolpath which maximizes the number of consecutive layers pr
inted within the same “island” without requiring any travel movement. In a
ddition, the algorithm optimizes the order of the printed islands, and gua
rantees that the generated toolpath will not cause collisions between the
extended nozzle and the model. We demonstrate our approach on a collection
of varied models, and show that we considerably reduce the required trav
el, and improve the printing time and quality compared to standard layer-b
ylayer printing.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 401
UID:123se24012024102440
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20220608T113000
DTEND;TZID="Asia/Jerusalem":20220608T123000
DTSTAMP;TZID="Asia/Jerusalem":20220608T113000
FREEBUSY;FBTYPE=BUSY:20220608T113000/20220608T123000
SUMMARY;LANGUAGE=en-US:ceClub talk by Oskar Mencer (Maxeler Technologies)
about ceClub: Conflict and Technology, AI Chip Wars from the Inside Out at
2022-06-08 11:30:00
DESCRIPTION;LANGUAGE=en-US:The microprocessor is 50 years old. 50 years ag
o, a single ALU at kHz speeds had to be shared by multiple tasks, multiple
applications, multiple users and multiple organizations. Due to transisto
r scaling we can now have 1M ALUs on a chip at GHz speeds. In this talk I
will swap the cause and effect equation, instead of talking about solution
s to problems, I will talk about problems created by solutions. Today data
movement dominates compute time. I will describe how by optimally solving
the data flow challenge, we are introducing a conflict with the past, the
present and the future; conflicts with science, technology, legal and bus
iness structures; conflicts with FLOPS, complexity theory, and supply chai
ns. Most recently, AI chips started forming the backbone of a concerted ef
fort around the world, to break the dead end position of the classical com
puting industry, and show multiple ways forward.
Bio: Oskar Mencer is
CEO of Maxeler Technologies, which has been recently acquired by Groq, an
AI chip company, to join forces in building the future of computing. Oska
r is a Technion graduate who grew up in the Parallel Systems Lab, class of
‘94, went on to Stanford, the Unix group at Bell Labs, Imperial College L
ondon, and more recently, became a member of Academia Europaea, and the Te
chnion Board of Governors.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Room 861, EE Meyer Building and zoom Lecture: 91011338796
UID:123se24012024102480
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20220608T123000
DTEND;TZID="Asia/Jerusalem":20220608T133000
DTSTAMP;TZID="Asia/Jerusalem":20220608T123000
FREEBUSY;FBTYPE=BUSY:20220608T123000/20220608T133000
SUMMARY;LANGUAGE=en-US:Theory Seminar talk by Itzhak Tamo (Tel-Aviv Univer
sity) about Theory Seminar: Nonlinear Repair Schemes of Reed-Solomon Codes
at 2022-06-08 12:30:00
DESCRIPTION;LANGUAGE=en-US:The problem of repairing linear codes, particul
arly Reed Solomon (RS) codes, has attracted a lot of attention in recent y
ears due to its importance in distributed storage systems. In this problem
, a failed code symbol (node) needs to be repaired by downloading as littl
e information as possible from a subset of the remaining nodes. There are
examples of RS codes with efficient repair schemes, and some are even opti
mal. However, these schemes fall short in several aspects; for example, th
ey require a considerable field extension degree, and in particular, they
do not work over prime fields. In this work, we explore the power of nonli
near repair schemes of RS codes and show that such schemes are crucial ove
r prime fields, and in some cases, they outperform all linear schemes.
Based on joint work with Roni Con.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 201
UID:123se24012024102520
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20220608T123000
DTEND;TZID="Asia/Jerusalem":20220608T143000
DTSTAMP;TZID="Asia/Jerusalem":20220608T123000
FREEBUSY;FBTYPE=BUSY:20220608T123000/20220608T143000
SUMMARY;LANGUAGE=en-US:CSpecial Event about Recruitment Day by Vast Data a
t 2022-06-08 12:30:00
DESCRIPTION;LANGUAGE=en-US:Vast Data engineers and recruitment teams will
arrive at CS to demonstrate technologies and off
er open positions<
/a>, on Wednesday, June 8, 2022, 12:30, in the Taub Lobby, and at 13:30 for a technological lecture on the challenges of b
uilding data structures and distributed algorith
ms used to establish the largest storage systems in the world, in the audi
torium of the visitor center (in front of the piano), on the entrance floo
r of the CS Taub Build
ing.
Please pre-register for the lecture (No obligation to submit CV
).
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:CS Taub Lobby and Visitors Center Auditorium
UID:123se24012024102540
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20220612T163000
DTEND;TZID="Asia/Jerusalem":20220612T203000
DTSTAMP;TZID="Asia/Jerusalem":20220612T163000
FREEBUSY;FBTYPE=BUSY:20220612T163000/20220612T203000
SUMMARY;LANGUAGE=en-US:CSpecial Event about TRX'22: Technion Robotics Expo
at 2022-06-12 16:30:00
DESCRIPTION;LANGUAGE=en-US:You are invited to the TRX'22: Tech
nion Robotics Expo, on Sunday, June 12, 2022, 16:30-20:00 in CS Taub A
uditorium 1.
Please pre-register.
More details and program.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:CS Taub Build. Auditorium 1
UID:123se24012024102580
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20220614T183000
DTEND;TZID="Asia/Jerusalem":20220614T203000
DTSTAMP;TZID="Asia/Jerusalem":20220614T183000
FREEBUSY;FBTYPE=BUSY:20220614T183000/20220614T203000
SUMMARY;LANGUAGE=en-US:CSpecial Event about Machine Learning and Real-time
AI: Reveal the Mystery at 2022-06-14 18:30:00
DESCRIPTION;LANGUAGE=en-US:You are invited to a technology enrichment lect
ure, by Nava Levy, AI / ML Dev Advocate, Redis, on machine learning in the
high-tech industry, with an emphasis on real-time machine learning, and t
he difference between it and deep learning and the challenges in real-time
AI, on roles and responsibilities of a typical data science and machine l
earning team in an industrial company, on the differences between research
and practice, and on tools that can be started today.
The lecture wi
ll be by zoom on Tuesday, June 14, 18:30, and a link to the meeting
will be sent after pre
-registration.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Event: Registration
UID:123se24012024102600
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20220615T123000
DTEND;TZID="Asia/Jerusalem":20220615T133000
DTSTAMP;TZID="Asia/Jerusalem":20220615T123000
FREEBUSY;FBTYPE=BUSY:20220615T123000/20220615T133000
SUMMARY;LANGUAGE=en-US:Theory Seminar talk by Klim Efremenko (Ben-Gurion U
niversity) about Theory Seminar: Binary Codes with Resilience Beyond 1/4 v
ia Interaction at 2022-06-15 12:30:00
DESCRIPTION;LANGUAGE=en-US: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 t
o encode x using an error-correcting code. As it is long known that the di
stance of binary codes is at most 1/2, reliable transmission is possible o
nly if the channel corrupts (flips) at most a 1/4-fraction of the communic
ated bits.
We revisit the reliable transmission problem in the two-wa
y 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 resi
lient to a constant fraction of corruptions strictly larger than 1/4. More
over, our code has a constant rate and requires Bob to only send one short
message.
Curiously, our new two-way code requires a fresh perspectiv
e 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 depen
ds on the “compatibility” of the messages they encode. We also prove that
such codes are necessary for our result.
Joint work with Gillat Kol,
Raghuvansh R. Saxena and Zhijun Zhang
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 201
UID:123se24012024102560
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20220619T193000
DTEND;TZID="Asia/Jerusalem":20220619T223000
DTSTAMP;TZID="Asia/Jerusalem":20220619T193000
FREEBUSY;FBTYPE=BUSY:20220619T193000/20220619T223000
SUMMARY;LANGUAGE=en-US:CSpecial Event about SHE S Ladies Night Eevent 2022
at 2022-06-19 19:30:00
DESCRIPTION;LANGUAGE=en-US:You are invited to the the lady students commun
ity annual event at Technion CS, on Sunday, June 19, 2022, starting at 19:
00 on the Taub terrace.
In the program:
19:00 - Mingling, sushi and
beers
20:00 - Stand-up show by Mor Chen
21:00 - Afterparty Karaoke
Please pre-
register (registration for the event involves a nominal fee).
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:CS Taub Terrace
UID:123se24012024102620
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20220620T123000
DTEND;TZID="Asia/Jerusalem":20220620T133000
DTSTAMP;TZID="Asia/Jerusalem":20220620T123000
FREEBUSY;FBTYPE=BUSY:20220620T123000/20220620T133000
SUMMARY;LANGUAGE=en-US:cggc talk by Prof. Justin Solomon (MIT) about CGGC
Seminar: Application-Driven Geometric Machine Learning at 2022-06-20 12:30
:00
DESCRIPTION;LANGUAGE=en-US:From 3D modeling to autonomous driving, a varie
ty of applications can benefit from data-driven reasoning about geometric
problems. The available data and preferred shape representation, however
, varies widely from one application to the next. Indeed, the one commona
lity among most of these settings is that they are not easily approached u
sing data-driven methods that have become de rigueur in other branches of
computer vision and machine learning. In this talk, I will summarize recen
t efforts in my group to develop learning architectures and methodologies
paired to specific applications, from point cloud processing to mesh and i
mplicit surface modeling. In each case, we will see how mathematical stru
ctures and application-specific demands drive our design of the learning m
ethodology, rather than bending application details or eliding geometric d
etails to apply a standard data analysis technique.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 401
UID:123se24012024102590
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20220621T113000
DTEND;TZID="Asia/Jerusalem":20220621T123000
DTSTAMP;TZID="Asia/Jerusalem":20220621T113000
FREEBUSY;FBTYPE=BUSY:20220621T113000/20220621T123000
SUMMARY;LANGUAGE=en-US:pixel-club talk by Amit Bermano (Tel-Aviv Universit
y) about Pixel Club: CLIP as a Generative Foundation Model at 2022-06-21
11:30:00
DESCRIPTION;LANGUAGE=en-US:Large scale Mega-models are impossible to train
using standard hardware, but encompass a vast semantic understanding of o
ur world. In this talk, I explore three ways to leverage the knowledge enc
ompassed in the Recent Large scale"Contrastive Language-Image Pre-training
" (CLIP) model, as a foundation to push the boundaries of generative capab
ilities:
- InStyleGAN-NADA, we show how to adapt the StyleGAN genera
tor across a multitudeof domains characterized by diverse styles and shape
s. Notably, the native editing capabilities StyleGAN offers are preserved
during this adaptation, and can be readily applied to the new domain.
-
InClipasso, we leverage only the visual (yet abstract and semantic) knowl
edge encoded in CLIP, to guide an object sketching optimization. The work
converts an image of an object to a sketch, allowing varying levels of abs
traction while preserving key visual features.
- InMotionClip, we intro
duce a 3D human motion auto-encoder featuring a latent embedding that is d
isentangled, well behaved, and supports highly semantic textual descriptio
ns, all thanks to CLIP. Specifically, we show how aligning the human motio
n manifold to CLIP space implicitly infuses the extremely rich semantic kn
owledge of CLIP into the manifold, even though it has not seen any tempora
l data during training.
Bio: Dr. Amit H. Bermano is a senior lectur
er (assistant professor) at the Blavatnik School of Computer Science in Te
l-Aviv University since 2018. His research focuses on Computer Graphics, C
omputer Vision, and Computational Fabrication, using tools from Machine Le
arning and Geometry Processing. Previously, he was a postdoctoral research
er at the Princeton Graphics Group, and a postdoctoral researcher at Disne
y Research Zurich. He has conducted his doctoral studies at ETH Zurich, in
collaboration with the computational materials group of Disney Research Z
urich. His Master’s and Bachelors’s degrees were obtained at the Technion
- Israel Institute of Technology
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Room 1061, EE Meyer Building
UID:123se24012024102640
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20220622T123000
DTEND;TZID="Asia/Jerusalem":20220622T133000
DTSTAMP;TZID="Asia/Jerusalem":20220622T123000
FREEBUSY;FBTYPE=BUSY:20220622T123000/20220622T133000
SUMMARY;LANGUAGE=en-US:Theory Seminar talk by Artur Ryazanov (St. Petersbu
rg University) about Theory Seminar: Communication Complexity-based Lower
Bounds for Proof Complexity of Natural Formulas at 2022-06-22 12:30:00
DESCRIPTION;LANGUAGE=en-US:A canonical communication problem Search(ϕ) is
defined for every unsatisfiable CNF ϕ: an assignment to the variables of ϕ
is distributed among the communicating parties, they are to find a clause
of ϕ falsified by this assignment. Lower bounds on the randomized k-party
communication complexity of Search(ϕ) in the number-on-forehead (NOF) mod
el imply tree-size lower bounds, rank lower bounds, and size-space tradeof
fs for the formula ϕ in the semantic proof system Tcc(k,c) that operates w
ith proof lines that can be computed by k-party randomized communication p
rotocol using at most c bits of communication [Göös, Pitassi, 2014]. All k
nown lower bounds on Search(ϕ) (e.g. [Impagliazzo, Pitassi, Urquhart, 1994
]; [Beame, Pitassi, Segerlind, 2007]; [Göös, Pitassi, 2014], ) are realize
d on ad-hoc formulas ϕ (i.e. they were introduced specifically for these l
ower bounds). We introduce a new communication complexity approach that al
lows establishing proof complexity lower bounds for natural formulas.
First, we demonstrate our approach for two-party communication and apply
it to the proof system Res(⊕) that operates with disjunctions of linear eq
ualities over F2 [Itsykson, Sokolov, 2014]. Let a formula PMG encode that
a graph G has a perfect matching. If G has an odd number of vertices, then
PMG has a tree-like Res (⊕)-refutation of a polynomial-size [Itsykson, So
kolov, 2014]. It was unknown whether this is the case for graphs with an e
ven number of vertices. Using our approach we resolve this question and sh
ow a lower bound 2Ω(n) on the size of tree-like Res(⊕)-refutations of PMKn
+2,n.
Then we apply our approach for k-party communication complexity
in the NOF model and obtain a Ω(1/k 2n/2k – 3k/2) lower bound on the rand
omized k-party communication complexity of Search BPHPM2n w.r.t. to some n
atural partition of the variables, where BPHPM2n is the bit pigeonhole pri
nciple and M=2n+2n(1-1/k). In particular, our result implies that the bit
pigeonhole requires exponential tree-like Th(k) proofs, where Th(k) is the
semantic proof system operating with polynomial inequalities of degree at
most k and k= O(log1-ε n) for some ϵ > 0. We also show that BPHP2n+12n su
perpolynomially separates tree-like Th(log1-ε m) from tree-like Th(log m),
where m is the number of variables in the refuted formula.
The talk
is based on joint work with Dmitry Itsykson.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 201
UID:123se24012024102630
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20220623T123000
DTEND;TZID="Asia/Jerusalem":20220623T143000
DTSTAMP;TZID="Asia/Jerusalem":20220623T123000
FREEBUSY;FBTYPE=BUSY:20220623T123000/20220623T143000
SUMMARY;LANGUAGE=en-US:CSpecial Talk talk by Prof. Adi Shamir (Weizmann In
stitute of Science) about Special Seminar: Efficient Detection of High Pro
bability Cryptographic Properties of Large Boolean Functions via Surrogate
Differentiation at 2022-06-23 12:30:00
DESCRIPTION;LANGUAGE=en-US:A central problem in cryptanalysis is to find a
ll the significant deviations from randomness in a given $n$-bit cryptogra
phic primitive. When $n$ is large, the only practical way to find such sta
tistical properties was to exploit the internal structure of the primitive
and to speed up the search with a variety of heuristic rules of thumb. Ho
wever, such bottom-up techniques can miss many properties, especially in c
ryptosystems which are designed to have hidden trapdoors.
In this tal
k I will consider the top-down version of the problem in which the cryptog
raphic primitive is given as a structureless black box which implements an
arbitrary Boolean function from $n$ bits to $n$ bits. I will then show ho
w to reduce the complexity of the best known techniques for finding all it
s significant differential and linear properties by a large factor of $2^{
n/2}$. The main new idea is to use {\it surrogate differentiation}, which
is a new way to analyze the properties of large Boolean functions. In the
context of finding differential properties, it enables us to simultaneousl
y find information about all the differentials of the form $f(x) \oplus f(
x \oplus \alpha)$ in all possible directions $\alpha$ by differentiating $
f$ in a single arbitrarily chosen direction $\gamma$ (which is unrelated t
o the $\alpha$'s). In the context of finding linear properties, surrogate
differentiation can be combined in a highly effective way with the Fast Fo
urier Transform.
This is joint work with Itai Dinur, Orr Dunkelman,
Nathan Keller, and Eyal Ronen.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:CS Taub Build. Auditorium 1
UID:123se24012024102550
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20220626T140000
DTEND;TZID="Asia/Jerusalem":20220626T160000
DTSTAMP;TZID="Asia/Jerusalem":20220626T140000
FREEBUSY;FBTYPE=BUSY:20220626T140000/20220626T160000
SUMMARY;LANGUAGE=en-US:CSpecial Event about Projects Fair on IoT, Android,
Arduino and Networks at 2022-06-26 14:00:00
DESCRIPTION;LANGUAGE=en-US:You are invited to the CS Taub projects fair fo
r the Spring Semester of 2022, where 30 teams of undergraduate students wi
ll present and demonstrate projects in various fields in IoT, Android, Ar
duino 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, a
nd were intended to make a contribution to the community.
The fair wi
ll be held on Sunday, June 26, 2022, 14:00-16:00, at the CS Taub Lobby.
The presenting posters (Heb)
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:CS Taub Lobby
UID:123se24012024102650
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20220626T163000
DTEND;TZID="Asia/Jerusalem":20220626T173000
DTSTAMP;TZID="Asia/Jerusalem":20220626T163000
FREEBUSY;FBTYPE=BUSY:20220626T163000/20220626T173000
SUMMARY;LANGUAGE=en-US:msc talk by Ron Marcovich about Protocol Infernece
from Program Executable Using Symbolic Execution and Automata Learning at
2022-06-26 16:30:00
DESCRIPTION;LANGUAGE=en-US:Protocol Inference is the process of gaining in
formation about a protocol from a binary code that implements it. This pro
cess is useful in cases such as extraction of the command and control prot
ocol of a malware, uncovering security vulnerabilities in a network protoc
ol implementation or verifying conformance to the protocol's standard. Pro
tocol inference usually involves time-consuming work to manually reverse e
ngineer the binary code.
We present a novel method to automatically i
nfer state machine of a network protocol and its message formats directly
from the binary code. To the best of our knowledge, this is the first meth
od to achieve this solely based on a binary code of a single peer. We do n
ot assume any of the following: access to a remote peer, access to capture
s of the protocol's traffic, and prior knowledge of message formats. The m
ethod leverages extensions to symbolic execution and novel modifications t
o automata learning. We validate the proposed method by inferring real-wor
ld protocols including the C&C protocol of Gh0st RAT, a well-known malware
.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture: 5088919659
UID:123se24012024102510
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20220627T133000
DTEND;TZID="Asia/Jerusalem":20220627T143000
DTSTAMP;TZID="Asia/Jerusalem":20220627T133000
FREEBUSY;FBTYPE=BUSY:20220627T133000/20220627T143000
SUMMARY;LANGUAGE=en-US:msc talk by Matan Mamistvalov about A New Lower Bou
nd on the Growth Constant of Polycubes in Three Dimensions at 2022-06-27 1
3:30:00
DESCRIPTION;LANGUAGE=en-US:In this thesis, we deal the approximation probl
em of the growth rate of polycubes in three dimensions. We consider three-
dimensional polycubes, which are finite collections of face-connected 3D-c
ubes, centered in points of $\mathbb{Z}^3$, where the lexicographically sm
allest cube is centered in $(0,0,0)$. If we denote the number of 3D polycu
bes comprised of $n$ cubes by $A(n)$, then we know from prior results that
this sequence behaves like an exponential, and so we denote its growth ra
te by $\lambda$. There are several prior lower bounds for $\lambda$, each
improving upon the other, the latest one before this work is by Barequet,
Ben-Shachar and Osegueda. In their work, the new method of \emph{recursive
compositions} is used explicitly, where it was implicitly in used in a fe
w previous works. In its essence, it creates distinct classes of polycubes
of size $n$ made of compositions of polycubes of smaller numbers,such tha
t when summing their cardinalities parametrized by $n$, we get lower bound
s on $A(n)$, which can then be used to get lower bounds on $\lambda$. They
also used the method of recursive compositions in a general manner fittin
g many multidimensional cubical lattices, achieving the lower bound of $6.
6521$. We make a more specialized use of this method to the 3D cubical lat
tice, achieving a lower bound of $\lambda \geq 6.6799$.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture: 98302771831
UID:123se24012024102450
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20220627T140000
DTEND;TZID="Asia/Jerusalem":20220627T150000
DTSTAMP;TZID="Asia/Jerusalem":20220627T140000
FREEBUSY;FBTYPE=BUSY:20220627T140000/20220627T150000
SUMMARY;LANGUAGE=en-US:msc talk by Barak Gahtan about Deep Reinforcement L
earning for 5G Dense Urban Wireless Routing at 2022-06-27 14:00:00
DESCRIPTION;LANGUAGE=en-US:We study the problem of routing real-time flows
over a multi-hop mmWave mesh. We develop a model-free Deep Reinforcement
Learning algorithm that determines which subset of the mmWave links should
be activated during each time slot and using what power level. The propos
ed algorithm, called AARL (Adaptive Activator RL), can handle a variety of
network topologies, packet loads and interference models. It does not req
uire prior knowledge of the interdependence of different mmWave links or t
he topology, and it is capable of adapting to different topologies. We de
monstrate AARL on three different topologies: a small topology with 10 lin
ks, a moderate topology with 48 links, and a large topology with 96 links.
We compare the results of AARL to those of a greedy algorithm. AARL is sh
own to outperform the greedy algorithm in two aspects. First, its schedule
is better since it obtains higher goodput. Second, and even more importan
tly, the running time of the greedy algorithm renders it impractical for r
eal-time scheduling, whereas the running time of AARL, is suitable for mee
ting the time constraints of typical 5g networks.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture: 94134094787
UID:123se24012024102530
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20220629T113000
DTEND;TZID="Asia/Jerusalem":20220629T123000
DTSTAMP;TZID="Asia/Jerusalem":20220629T113000
FREEBUSY;FBTYPE=BUSY:20220629T113000/20220629T123000
SUMMARY;LANGUAGE=en-US:pixel-club talk by Maxim Fishman (EE, Technion) abo
ut Pixel Club: Graph Neural Networks through the Lens of Measure Theory at
2022-06-29 11:30:00
DESCRIPTION;LANGUAGE=en-US:Despite their growing popularity, graph neural
networks (GNNs) still suffer from multiple unsolved problems, including la
ck of embedding expressiveness, propagation of information to distant node
s, and training on large-scale graphs. Understanding the roots of and prov
iding solutions for such problems require developing analytic tools and te
chniques. In this talk we provide a measure theoretic point of view for t
he above-mentioned problems, and derive a notion of “recoverability” which
will serve us as a tool for GNN embedding analysis, unsupervised graph re
presentation learning and regularization. At the end of the talk, we will
show a tight relationship between recoverability loss minimization and mut
ual information maximization.
M.Sc. student under the supervision of
Prof. Avi Mendelson and Dr. Chaim Baskin.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture: 4108205267
UID:123se24012024102680
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20220629T123000
DTEND;TZID="Asia/Jerusalem":20220629T133000
DTSTAMP;TZID="Asia/Jerusalem":20220629T123000
FREEBUSY;FBTYPE=BUSY:20220629T123000/20220629T133000
SUMMARY;LANGUAGE=en-US:Theory Seminar talk by Amir Abboud (Weizmann Instit
ute of Science) - CANCELLED! about Theory Seminar: APMF < APSP? Gomory-
Hu Tree in Subcubic Time at 2022-06-29 12:30:00
DESCRIPTION;LANGUAGE=en-US:The All-Pairs Max-Flow problem (APMF) asks to c
ompute the maximum flow (or equivalently, the minimum cut) between all pai
rs of nodes in a graph. The naive solution of making n^2 calls to a (singl
e-pair) max-flow algorithm was beaten in 1961 by a remarkable algorithm of
Gomory and Hu that only makes n-1 calls. Within the same time bound, thei
r algorithm also produces a cut-equivalent tree (a.k.a. GH-Tree) that pres
erves all pairwise minimum cuts exactly. This gives a cubic upper bound fo
r APMF assuming that single-pair max-flow can be solved optimally and the
only improvements since 1961 have been on getting us closer to this assump
tion; new algorithms that break the cubic barrier were only known for spec
ial graph classes or with approximations.
The All-Pairs Shortest-Path
s problem (APSP) is similar, but asks to compute the distance rather than
the connectivity between all pairs of nodes. Its time complexity also appe
ars similar, with classical cubic time algorithms that have only been brok
en in special cases or with approximations. Meanwhile, in the past 10 year
s, the conjecture that APSP requires cubic time has played a central role
in fine-grained complexity, leading to cubic conditional lower bounds for
many other fundamental problems that appear even easier than APMF. However
, a formal reduction from APSP to APMF has remained elusive.
This tal
k will survey recent progress (based on joint works with Robert Krauthgame
r and Ohad Trabelsi), starting with partially successful attempts at reduc
ing APSP to APMF, going through algorithmic progress on APMF in limited se
ttings, and leading up to a very recent paper (also with Jason Li, Debmaly
a Panigrahi, and Thatchaphol Saranurak) where we break the 60-year old cub
ic barrier for APMF; suggesting a separation between APMF and APSP.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 201
UID:123se24012024102660
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20220629T163000
DTEND;TZID="Asia/Jerusalem":20220629T173000
DTSTAMP;TZID="Asia/Jerusalem":20220629T163000
FREEBUSY;FBTYPE=BUSY:20220629T163000/20220629T173000
SUMMARY;LANGUAGE=en-US:phd talk by Gali Sheffi about Reliable Concurrent C
omputing at 2022-06-29 16:30:00
DESCRIPTION;LANGUAGE=en-US:The rapid deployment of multi-core architecture
s has resulted in a dire need for scalable and reliable concurrent algorit
hms. This dissertation focuses on the design of concurrent data structures
, which constitute building blocks for concurrent algorithms. Two major de
sign goals in this domain are reliability and efficiency. This talk will c
oncentrate on the hardness of reclaiming concurrent data-structures' memor
y efficiently, while taking care to preserve reliability. It will include
(1) the design of a novel and efficient concurrent reclamation scheme, (2)
its application to multi version concurrency control (i.e., range queries
), and (3) a general impossibility demonstrating the hardness of designing
reliable memory reclamation schemes.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture: 95272009519
UID:123se24012024102570
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20220630T113000
DTEND;TZID="Asia/Jerusalem":20220630T123000
DTSTAMP;TZID="Asia/Jerusalem":20220630T113000
FREEBUSY;FBTYPE=BUSY:20220630T113000/20220630T123000
SUMMARY;LANGUAGE=en-US:msc talk by Shai Guendelman about Concurrent Games
with Multiple Topologies at 2022-06-30 11:30:00
DESCRIPTION;LANGUAGE=en-US:Concurrent multi-player games with omega-regula
r objectives are a standard model for systems that consist of several inte
racting components, each with its own objective. The standard solution con
cept for such games is Nash Equilibrium (NE), which is a ``stable'' strate
gy profile for the players. In many settings, the system is not fully obse
rvable by the interacting components, e.g., due to internal variables. The
n, the interaction is modelled by a partial information game. Unfortunatel
y, the problem of whether a partial information game has an NE is not kno
wn to be decidable. A particular setting of partial information arises nat
urally when processes are assigned IDs by the system, but these IDs are no
t known to the processes. Then, the processes have full information about
the state of the system, but are uncertain of the effect of their actions
on the transitions. We generalize the setting above and introduce Multi-To
pology Games (MTGs) -- concurrent games with several possible topologies,
where the players do not know which topology is actually used. We show tha
t extending the concept of NE to these games can take several forms. To th
is end, we propose two notions of NE: Conservative NE, in which a player d
eviates if she can strictly add topologies to her winning set, and Greedy
NE, where she deviates if she can win in a previously-losing topology. We
study the properties of these NE, and show that the problem of whether a g
ame admits them is decidable.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture: 94035960574
UID:123se24012024102610
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20220719T130000
DTEND;TZID="Asia/Jerusalem":20220719T140000
DTSTAMP;TZID="Asia/Jerusalem":20220719T130000
FREEBUSY;FBTYPE=BUSY:20220719T130000/20220719T140000
SUMMARY;LANGUAGE=en-US:phd talk by Ron Slossberg about On Synthesis and Re
construction of human Facial Photometry and Corresponding Geometry at 2022
-07-19 13:00:00
DESCRIPTION;LANGUAGE=en-US:In this thesis, we study the modeling of human
faces. As all structured data is believed to reside on some low-dimensiona
l manifold in a high-dimensional space, we wish to study and model the so-
called manifold of human faces. By uncovering the latent manifold of faces
one can project onto the manifold (facial reconstruction) as well as samp
le from the manifold (facial synthesis), two tasks with a wide range of ap
plications such as gaming, animation, and AR/VR to name a few.
In the
ir seminal work (1999), Vetter and Blanz proposed the linear 3D Morphable
Model (3DMM). This model has been widely adopted since and can be thought
of as a first-order linear approximation of the facial manifold. This mode
l, however, has two main drawbacks: The fine details are lost, and it is n
ot well suited for facial synthesis. The first drawback stems from the tru
ncation of the PCA basis as well as the linear nature of the model. The se
cond one arises since the proposed method for randomly selecting coefficie
nts for each vector does not consider the latent facial manifold.
In
our work, we wish to remedy these problems by constructing a non-linear m
odel for facial photometry and combining it with the linear geometric 3DMM
to achieve highly realistic facial modeling. Our approach leverages the G
enerative Adversarial Network (GANs ) training methodology to achieve a no
n-linear model for texture generation. To form the final 3D face, we propo
se methods for generating a corresponding geometry via the 3DMM model for
each synthesized texture. In addition, we can project onto the facial mani
fold by optimizing the generator input parameters according to some image
loss by leveraging backward propagation through the generator model. This
process enables us to perform a full facial reconstruction even under chal
lenging circumstances such as side-views.
Initially, we propose to le
arn the model in a supervised manner directly from facial scans. This is d
one by performing semantic alignment of the scans and mapping the scanned
textures to texture images used during the training process. We later prop
ose a new unsupervised methodology based only on natural facial photos. Th
is is much more practical and yields better results due to the much larger
dataset, however, training without direct supervision of textures is more
complicated and requires a complex training pipeline. Compared to previou
s efforts we are among the very few proposed methods for facial generation
and our results are shown to be SOTA in this task. For the task of facial
reconstruction via our model, we compete with many prior methods and demo
nstrate that we compare favorably, even outperforming several supervised m
ethods.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture: 99388657358
UID:123se24012024102700
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20220721T130000
DTEND;TZID="Asia/Jerusalem":20220721T140000
DTSTAMP;TZID="Asia/Jerusalem":20220721T130000
FREEBUSY;FBTYPE=BUSY:20220721T130000/20220721T140000
SUMMARY;LANGUAGE=en-US:msc talk by Amani Shhadi about Topology-Controlled
Reconstruction from Partial Planar Cross-Sections at 2022-07-21 13:00:00
DESCRIPTION;LANGUAGE=en-US:The problem of three-dimensional reconstruction
from planar cross-sections arises in many fields, such as biomedical imag
e analysis, and geographical information systems. The problem has been stu
died extensively in the past~40 years.
Each cross-section of the inpu
t contains multiple contours, where each contour divides the plane into di
fferent material types.
The reconstructed three-dimensional object i
s a valid volume (surrounded by a closed surface) that interpolates the in
put slices.
Some existing works utilize prior information about the r
econstructed object, such as its topology, which can be described by the n
umber of connected components and the genus of each component, for recover
ing the original shape of the reconstructed object. These works assume tha
t the input cross-sections are complete and do not contain missing informa
tion. In most real-life cases, this assumption does not hold, and the inpu
t cross-sections might contain noisy or unknown areas.
Other existi
ng works handle such inputs; however, these methods do not have topologica
l guarantees for the reconstructed object.
In this work, we provide
, to our best knowledge, the first algorithm that provides topology contro
l for three-dimensional reconstruction from partial planar cross-sections.
The input to our algorithm is an arbitrarily-oriented set of 2-dime
nsional cross-sections that might contain missing information, which we re
fer to as “unknown'' regions, in addition to user-specified topology const
raints on the reconstructed object.
During the reconstruction process
, we explore a set of distinct topologies for relabeling the “unknown'' re
gions. We define a scoring function for calculating the likelihood of each
topology.
We choose one set of topologies, so that the reconstructed
object satisfies the global topology constraints and the score function i
s maximized.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture: 9336114415 and Taub 301
UID:123se24012024102720
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20220724T110000
DTEND;TZID="Asia/Jerusalem":20220724T120000
DTSTAMP;TZID="Asia/Jerusalem":20220724T110000
FREEBUSY;FBTYPE=BUSY:20220724T110000/20220724T120000
SUMMARY;LANGUAGE=en-US:cggc talk by Hsueh-Ti Derek Liu (University of Toro
nto) about CGGC Seminar: 3D Content Creation Made Fast and Easy at 2022-07
-24 11:00:00
DESCRIPTION;LANGUAGE=en-US:Creating digital 3D objects has been a central
task across different disciplines and the key towards democratizing the me
taverse. However, 3D content creation is still a privilege reserved for pr
ofessional modelers because existing content creation tools are difficult
to use by the general public. My research aims to lower the difficulty of
3D content creation to the point where everyone can manipulate digital 3D
shapes. In this talk, I will first discuss how to build easy-to-use modeli
ng algorithms using machine learning. Specifically, this involves developi
ng network architectures on triangle meshes, which are robust to shape var
iants (e.g., different resolution and rigid transformation). We demonstrat
e how our approach can generalize even when trained on only a single shape
. I will then expand the discussion to numerical methods -- multigrid meth
ods on curved surfaces -- that are crucial to support interactive content
creation at scale. I believe that future 3D content creation will involve
more high-level and ``smart'' operations. Such operations will require rob
ust geometric learning to harness 3D data in the wild and numerical method
s to process geometric data at scale. By solving these challenges, I argue
that easy-to-use 3D content creation tools will push the boundaries of fa
brication, architecture, and democratize the metaverse.
Bio:
Hsu
eh-Ti Derek Liu is a Ph.D. candidate at the University of Toronto studying
digital geometry processing. Derek's work mainly focuses on developing ea
sy-to-use 3D modeling tools and scalable numerical methods for processing
geometric data. His research is published at top-tier venues (ACM SIGGRAPH
and ICLR) and is supported by an Adobe Research Fellowship. His PhD advis
or is Prof. Alec Jacobson. He worked as a visiting scholar at École Polyte
chnique in 2019, working with Prof. Maks Ovsjanikov. He completed his M.S.
with Profs. Keenan Crane and Levent Burak Kara at Carnegie Mellon Univers
ity. (Website - https://www.dgp.toronto.edu/~hsuehtil)
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Room 337 Taub Bld.
UID:123se24012024102670
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20220726T103000
DTEND;TZID="Asia/Jerusalem":20220726T113000
DTSTAMP;TZID="Asia/Jerusalem":20220726T103000
FREEBUSY;FBTYPE=BUSY:20220726T103000/20220726T113000
SUMMARY;LANGUAGE=en-US:msc talk by Linor Ackerman-Schraier about A Machine
Learning Exploration of Relations between Protein Structures and their Ge
netic Coding at 2022-07-26 10:30:00
DESCRIPTION;LANGUAGE=en-US:Synonymous codons translate into the same amino
acid. Although the identity of synonymous codons is often considered inco
nsequential to the final protein structure there is mounting evidence for
an association between the two. Protein structure plays an important role
in understanding the biological function and mechanism of a protein theref
ore understanding the relations between protein structures and their genet
ic coding is crucial. Our study examined the association between the two b
y using regression and classification models and found that (i) codon sequ
ences predict protein backbone dihedral angles with a lower error than ami
no acid sequences, and (ii) models trained with true dihedral angles have
better classification of synonymous codons given structural information th
an models trained with random dihedral angles. Using this classification a
pproach, we investigate the local codon-codon dependencies and test whethe
r synonymous codon identity can be predicted more accurately from codon co
ntext rather than amino acid context, and most specifically which codon co
ntext position carries the most predictive power.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture: 7665101196
UID:123se24012024102710
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20220731T090000
DTEND;TZID="Asia/Jerusalem":20220812T200000
DTSTAMP;TZID="Asia/Jerusalem":20220731T090000
FREEBUSY;FBTYPE=BUSY:20220731T090000/20220812T200000
SUMMARY;LANGUAGE=en-US:CSpecial Event about FLOC 2022: The Eighth Federate
d Logic Conference at 2022-07-31 09:00:00
DESCRIPTION;LANGUAGE=en-US:FLOC 2022: The Eighth Federated Logic Conferenc
e (FLoC 2022,July 31-August 12, 2022, Haifa, Israel)
Hosted by the He
nry and Marilyn Taub Faculty of Computer Science at the Technion
ABOU
T FLOC
During the past forty years there has been extensive, continuous,
and growing interaction between logic and computer science. In many respe
cts, logic provides computer science with both a unifying foundational fra
mework and a tool for modeling. In fact, logic has been called “the calcul
us of computer science”, playing a crucial role in diverse areas such as a
rtificial intelligence, computational complexity, distributed computing, d
atabase systems, hardware design, programming languages, and software engi
neering.
The Federated Logic Conference brings together several inter
national conferences related to mathematical logic and computer science an
d was first organized in 1996, as part of the DIMACS Special Year on Logic
and Algorithms. Since then FLoC was held in Trento in 1999, Copenhagen in
2002, Seattle in 2006, Edinburgh in 2010, Vienna in 2014, and Oxford in 2
018.
REGULAR REGISTRATION CLOSES on 20th July 2022.
ON-SITE REGI
STRATION will be possible during the conference.
Details, program and registration.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Technion, Haifa
UID:123se24012024102690
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20220810T123000
DTEND;TZID="Asia/Jerusalem":20220810T133000
DTSTAMP;TZID="Asia/Jerusalem":20220810T123000
FREEBUSY;FBTYPE=BUSY:20220810T123000/20220810T133000
SUMMARY;LANGUAGE=en-US:Theory Seminar talk by Seri Khoury (UC Berkeley) ab
out Theory Seminar: Hardness of Approximation in P via Short Cycle Removal
: Cycle Detection, Distance Oracles, and Beyond at 2022-08-10 12:30:00
DESCRIPTION;LANGUAGE=en-US:Triangle finding is at the base of many conditi
onal lower bounds in P, mainly for distance computation problems, and the
existence of many $4$- or $5$-cycles in a worst-case instance had been the
obstacle towards resolving major open questions.
We present a new te
chnique for efficiently removing almost all short cycles in a graph withou
t unintentionally removing its triangles. Consequently, triangle finding p
roblems do not become easy even in almost $k$-cycle free graphs, for any c
onstant $k\geq 4$. This allows us to establish new hardness of approximati
on results for distance related problems, such as distance oracles, dynami
c shortest paths, and more.
Based on a joint work with Amir Abboud, K
arl Bringmann, and Or Zamir (STOC 2022).
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Amado 814
UID:123se24012024102760
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20220811T130000
DTEND;TZID="Asia/Jerusalem":20220811T140000
DTSTAMP;TZID="Asia/Jerusalem":20220811T130000
FREEBUSY;FBTYPE=BUSY:20220811T130000/20220811T140000
SUMMARY;LANGUAGE=en-US:phd talk by Gil Ben-Shachar about Extremal Properti
es of Polyominoes at 2022-08-11 13:00:00
DESCRIPTION;LANGUAGE=en-US:Lattice animals are edge-connected sets of cell
s over various lattices. Some famous examples are polyominoes, polyhexes,
polyiamonds and polycubes which are in the square, hexagonal triangular an
d cubical lattices respectively. Lattice animals have been studied extensi
vely both as a combinatorial object, and as modeling tool in statistical p
hysics.
The two most significant problems related to lattice animals
are the counting problem and the growth rate problem. The first is simply
counting how many lattice animals with $n$ calls are there. This number is
denoted by $A_\lattice(n)$ where $\lattice$ describes the specific lattic
e in question. The second problem relates to the growth rate of $A_\lattic
e(n)$. It have been shown that $\lim_{n \to \infty} A_\lattice(n+1)/A_\lat
tice(n)$ exists for all major lattices, and is denoted by $\lambda_\lattic
e$. The exact values of $\lambda_\lattice$ are unknown, however some bound
s exists for it.
Another area that have seen extensive research is st
udying lattice animals by their perimeter. The perimeter of a lattice anim
al is the set of unoccupied cells adjacent to the cells of the animal. The
motivation to study polyominoes by their perimeter emerge from statistica
l physics where the numbers of lattice animals with a certain number of ce
lls and perimeter size are used to model statistical processes such as per
colation processes.
In this work we some questions related to polyomi
noes and lattice animals. We studied minimum-perimeter lattice animals, wh
ich are lattice animals with the minimum number of perimeter cells for the
ir area. We show some combinatorial properties of minimum-perimeter lattic
e animals, and by doing so proving a long standing conjecture in the field
of enumerative chemistry. We also provide efficient algorithms for the en
umeration of minimum-perimeter lattice animals.
We also studied the n
umber of ways two polyominoes and polycubes can be composed with each othe
r to create a new, larger, polycube. This question emerged from an older s
tudy which tried to improve the upper bound on the growth rate of $A_\latt
ice(n)$ for polyominoes.
Another area of study in this work was to st
udy a classic argument that was used for the calculation of bounds on $\la
mbda_\lattice$ for various lattices. We refine this argument and by doing
so we improved the best known bounds for the growth rate of certain classe
s of polyominoes and polycubes.
Lastly, we tackled one of the biggest
problems in the field of studying polyominoes, computing $A_\lattice(n)$
for the square lattice. We were able to improve on the best known algorit
hm and by doing so extend the known values of $A_\lattice(n)$ from $n=56$
to $n=70$.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture: 93608083202 and Taub 601
UID:123se24012024102730
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20220816T130000
DTEND;TZID="Asia/Jerusalem":20220816T140000
DTSTAMP;TZID="Asia/Jerusalem":20220816T130000
FREEBUSY;FBTYPE=BUSY:20220816T130000/20220816T140000
SUMMARY;LANGUAGE=en-US:phd talk by Dor Harris about On the Orchestration o
f Advanced Cellular Networks at 2022-08-16 13:00:00
DESCRIPTION;LANGUAGE=en-US:In recent years network operators are experienc
ing changes in clients needs. Self-driving cars, augmented reality games a
nd large scale data streaming are simple examples of new applications that
require faster service with higher bandwidth availability to the clients.
These changes force the network operators shift their business model and
new network paradigms arise.
The ongoing transition into 5G networks
(and 6G networks that will soon arrive) is enabled in part by the combinat
ion of NFV (Network Function Virtualization) and MEC (Multi-access Edge C
omputing), two promising paradigms that allow executing ultra-low-latency
network where services are located on edge nodes, physically closer to the
clients.
These two paradigms complete each other, the main idea be
hind NFV is decoupling functionality from hardware, by moving function fro
m hardware based server to being software-based and deployed on off-the-sh
elf commodity server, it allows networks to be more agile where service lo
cation may change swiftly. MEC, on the other hand, allows to move services
from centralized data centers within network's core to the network's edge
, which allows the network to provide service with lower latency due to sh
ort distance to the clients. Therefore, orchestrating this complex distrib
uted environment and especially provisioning services in a timely manner,
in order to address the dynamic workload, created a big challenge.
In
order to utilize these paradigms, networks operators needs to deploy them
efficiently by using new resource allocation algorithms. In this thesis w
e provide several algorithmic solutions that can help better utilize these
emerging networks. We define a rigorous model and present an algorithmic
solution for the specific problems. We provide analytically proven perform
ance bounds for these algorithms that are compared to the relevant lower b
ound. For some of the problems we also present a thorough performance eval
uation via extensive simulation, indicating their advantage over other sol
utions in realistic scenarios.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture: 92767988833
UID:123se24012024102750
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20220822T083000
DTEND;TZID="Asia/Jerusalem":20220822T093000
DTSTAMP;TZID="Asia/Jerusalem":20220822T083000
FREEBUSY;FBTYPE=BUSY:20220822T083000/20220822T093000
SUMMARY;LANGUAGE=en-US:msc talk by Roded Zats about Fair Correlation Clust
ering In General Graphs at 2022-08-22 08:30:00
DESCRIPTION;LANGUAGE=en-US:We consider the family of Correlation Clusterin
g optimization problems under fairness constraints.
In Correlation Clust
ering we are given a graph whose every edge is labeled either with a $+$ o
r a $-$, and the goal is to find a clustering that agrees the most with th
e labels: $+$ edges within clusters and $-$ edges across clusters.
The n
otion of fairness implies that there is no over, or under, representation
of vertices in the clustering: every vertex has a color and the distribut
ion of colors within each cluster is required to be the same as the distri
bution of colors in the input graph.
Previously, approximation algorithm
s were known only for fair disagreement minimization in complete unweighte
d graphs.
We prove the following:
$(1)$ there is no finite approxim
ation for fair disagreement minimization in general graphs unless $ P=NP$
(this hardness holds also for bicriteria algorithms); and $(2)$ fair agree
ment maximization in general graphs admits a bicriteria approximation of $
\approx 0.591$ (an improved $\approx 0.609$ true approximation is given fo
r the special case of two uniformly distributed colors).
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture: 7795921179
UID:123se24012024102740
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20220828T110000
DTEND;TZID="Asia/Jerusalem":20220828T120000
DTSTAMP;TZID="Asia/Jerusalem":20220828T110000
FREEBUSY;FBTYPE=BUSY:20220828T110000/20220828T120000
SUMMARY;LANGUAGE=en-US:msc talk by Roy Benjamin about Graph Neural Network
s Pretraining Through Inherent Supervision for Molecular Property Predicti
on at 2022-08-28 11:00:00
DESCRIPTION;LANGUAGE=en-US:Recent global events have emphasized the import
ance of accelerating the drug discovery process. This process may take mor
e than a decade and its overall cost might exceed one billion dollars. A w
ay to deal with these issues is to use machine learning to increase the ra
te at which drugs are made available to the public while reducing the cost
of the entire process. However, chemical labeled data for real-world appl
ications is extremely scarce making traditional approaches less effective.
A promising course of action for this challenge is to pretrain a mode
l using related tasks with large enough datasets, with the next step being
finetuning it for the desired task. This is challenging as creating these
datasets requires labeled data or expert knowledge.
To aid in solving
this pressing issue, in this thesis we introduce MISU - Molecular Inherent
SUpervision, a unique method for pretraining graph neural networks for mo
lecular property prediction.
Our method leapfrogs past the need for la
beled data or any expert knowledge by introducing three innovative compone
nts that utilize inherent properties of molecular graphs to induce informa
tion extraction at different scales, from the local neighborhood of an ato
m to substructures in the entire molecule.
We evaluate our framework o
n six chemical property prediction tasks. We show that our method reaches
state-of-the-art results compared to numerous baselines. We conduct a thor
ough ablation experiment and emphasize the contribution of each component
in the method. In addition, we explore the effect of MISU on various GNN a
rchitectures and find our method is consistent with work done on supervise
d pretraining.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture: 96304969082
UID:123se24012024102770
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20220904T110000
DTEND;TZID="Asia/Jerusalem":20220904T120000
DTSTAMP;TZID="Asia/Jerusalem":20220904T110000
FREEBUSY;FBTYPE=BUSY:20220904T110000/20220904T120000
SUMMARY;LANGUAGE=en-US:msc talk by Yakir Yehuda about A Framework for Clin
ical Classification of Multivariate Time Series using Koopman Operators at
2022-09-04 11:00:00
DESCRIPTION;LANGUAGE=en-US:Clinical multivariate time series (MTS) arising
from sensor data, such as EEG and ECG, is used in a variety of tasks.
The sensors are composed of multiple leads connected to the body, where
each lead generates a time series of data.
Combining information fr
om the different leads allows inference of cardiac activity from ECG (arrh
ythmia, acute coronary syndrome) or brain dysfunction from EEG (brain tumo
rs, strokes, epilepsy).
We present a framework for clinical classific
ation of MTS based on their joint dynamics. Specifically, we learn a dyna
mical system describing the evolution of the multiple signals together in
time based on the theory of Koopman operators.
According to Koopman t
heory, a high-dimensional embedding space exists in which the operator pro
pagating from one time instant to the next is linear; thus, we learn both
the mapping to this embedding space and the linear operator that correspon
ds to it. We then pose a joint optimization framework and learn the linear
MTS dynamics, while simultaneously optimizing the loss corresponding to t
he original classification task, where classification depends on the Koopm
an embedding.
Our technique yields reliable clinical diagnosis in an
empirical study employing signals from thousands of patients in multiple c
linical tasks employing two types of clinical-grade sensors (ECG and EEG).
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture: 2488049770
UID:123se24012024102780
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20220908T110000
DTEND;TZID="Asia/Jerusalem":20220908T120000
DTSTAMP;TZID="Asia/Jerusalem":20220908T110000
FREEBUSY;FBTYPE=BUSY:20220908T110000/20220908T120000
SUMMARY;LANGUAGE=en-US:msc talk by Adi Simhi about Interpreting Embedding
Spaces by Conceptualization at 2022-09-08 11:00:00
DESCRIPTION;LANGUAGE=en-US:One of the main methods for semantic interpreta
tion of text is mapping it into a vector in some embedding space. Such vec
tors can then be used for a variety of textual processing tasks. Recently
, most embedding spaces are a product of training large language models.
One major drawback of this type of representation is their incomprehensibi
lity to humans. Understanding the embedding space is crucial for several
important needs, including the need to explain the decision of a system th
at uses the embedding, the need to debug the embedding method and compare
it to alternatives, and the need to detect biases hidden in the model.
In this paper, we present a novel method of transforming any embedding s
pace to a comprehensible conceptual space. We first present an algorithm
for deriving a conceptual space with dynamic on-demand granularity. We th
en show a method for transferring any vector in the original incomprehensi
ble space to an understandable vector in the conceptual space. We combine
human tests with cross-model tests to show that the conceptualized vector
s indeed represent the semantics of the original vectors. We also show ho
w the conceptualized vectors can be used for various tasks including ident
ifying weaknesses in the semantics underlying the original spaces and diff
erences in the semantics of alternative models.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture: 99807656912
UID:123se24012024102800
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20220908T133000
DTEND;TZID="Asia/Jerusalem":20220908T143000
DTSTAMP;TZID="Asia/Jerusalem":20220908T133000
FREEBUSY;FBTYPE=BUSY:20220908T133000/20220908T143000
SUMMARY;LANGUAGE=en-US:msc talk by Antoine Vinciguerra about On The Struct
ure Of Heilbronn’s Configurations at 2022-09-08 13:30:00
DESCRIPTION;LANGUAGE=en-US:Heilbronn's triangle problem asks how to place
n points in the unit square, such that the smallest of the $\binom{n]{ 3}$
triangles is maximized. This problem, which was opened in the 1950s by th
e mathematician Heilbronn, has not yet been answered for n>8. Calling H)n(
the area of the smallest triangle of the best set of n points, no tight b
ounds on it have been found so far. The best upper and lower bounds are $O
(n^{-\mu+\epsilon})$, where $\mu=\frac{8}{7}$ , and $\Omega(\frac{\log n}
{n^2})$ respectively, which still leaves a large gap. Heilbronn's triangle
problem has many variants, for example, reducing the set of possible poin
t sets to those where some properties hold. On the other hand, finding a
'good' solution to Heilbronn's problem has been attempted algorithmically
up to only 24 points since the complexity of the algorithms used making th
em non-applicable for large number of points.
In fact, not much has b
een found about the structure of the best point sets, such as the number o
f smallest triangles, the number of points lying on the boundary of the un
it square, etc. In this talk, we investigate the location of points in lo
cally-optimum configurations of points, as well as some properties that sh
ould hold for optimal point sets. From those properties, we also derive ot
her properties for variants of Heilbronn's triangle problem, in which the
point are in monotone-decreasing position or convex-decreasing position. F
inally, we use a particle swamp algorithm variation in order to find a sol
ution as good as possible for $n>24$, thus, reducing time complexity and b
eing as close as possible to the best known point sets for $n\leq 24$.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture: 96677350253
UID:123se24012024102790
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20220912T140000
DTEND;TZID="Asia/Jerusalem":20220912T150000
DTSTAMP;TZID="Asia/Jerusalem":20220912T140000
FREEBUSY;FBTYPE=BUSY:20220912T140000/20220912T150000
SUMMARY;LANGUAGE=en-US:msc talk by Shafik Nassar about Succinct Interactiv
e Oracle Proofs at 2022-09-12 14:00:00
DESCRIPTION;LANGUAGE=en-US:\textit{Interactive Oracle Proofs} (IOPs) are a
new type of proof-system that combines key properties of interactive proo
fs and PCPs: IOPs enable a verifier to be convinced of the correctness of
a statement by interacting with an untrusted prover while reading just a f
ew bits of the messages sent by the prover. IOPs have become very prominen
t in the design of efficient proof-systems in recent years.
In this
work we study \textit{succinct IOPs}, which are IOPs in which the communic
ation complexity is polynomial (or even linear) in the original witness. W
hile there are strong impossibility results for the existence of succinct
PCPs (i.e., PCPs whose length is polynomial in the witness), it is known t
hat the rich class of NP relations that are decidable in small space have
succinct IOPs. In this work we show both new applications, and limitations
, for succinct IOPs:
First, using one-way functions, we show how to
compile IOPs into zero-knowledge \textit{proofs}, while nearly preserving
the proof length. This complements a recent line of work, initiated by Ben
-Sasson et al. (TCC, 2016B), who compile IOPs into super-succinct zero-kno
wledge \textit{arguments}.
Applying the compiler to the state-of-the
-art succinct IOPs yields zero-knowledge proofs for bounded-space NP relat
ions, with communication that is nearly equal to the original witness leng
th. This yields the shortest known zero-knowledge proofs from the minimal
assumption of one-way functions.
Second, we give a barrier for obtai
ning succinct IOPs for more general NP relations. In particular, we show t
hat if a language has a succinct IOP, then it can be decided in \textit{sp
ace} that is proportionate only to the witness length, after a bounded-tim
e probabilistic preprocessing. We use this result to show that under a sim
ple and plausible (but to the best of our knowledge, new) complexity-theor
etic conjecture, there is no succinct IOP for CSAT.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture: 3915434332 and Taub 601
UID:123se24012024102820
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20220921T103000
DTEND;TZID="Asia/Jerusalem":20220921T113000
DTSTAMP;TZID="Asia/Jerusalem":20220921T103000
FREEBUSY;FBTYPE=BUSY:20220921T103000/20220921T113000
SUMMARY;LANGUAGE=en-US:msc talk by Guy Azran about Generalizing Reinforcem
ent Learning Agents with Abstract Contextual Embeddings at 2022-09-21 10:3
0:00
DESCRIPTION;LANGUAGE=en-US:Classic reinforcement learning methods such as
Q-learning and policy gradient methods have seen great success in learning
to perform a plethora of common tasks in AI, from playing video games to
controlling self-driving vehicles and beyond. As a result, these methods a
nd their extensions have become standard in most reinforcement learning se
ttings. However, they have trouble adapting to changes in the environment.
We believe this issue can be solved by giving the agent awareness of thes
e changes in the form of context that could potentially steer it toward an
optimal policy regardless of the nature of these changes. In this proposa
l, we suggest a novel learning method that provides said context in the fo
rm of a graph abstraction of the current environment to the agent. We do t
his by modeling the contexts as reward machines and using graph neural net
works to embed them into meaningful representations that an agent can leve
rage to understand the current environment status.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture: 99606696165
UID:123se24012024102810
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20220928T110000
DTEND;TZID="Asia/Jerusalem":20220928T120000
DTSTAMP;TZID="Asia/Jerusalem":20220928T110000
FREEBUSY;FBTYPE=BUSY:20220928T110000/20220928T120000
SUMMARY;LANGUAGE=en-US:msc talk by Avigail Cohen-Rimon about Geometry-base
d Dynamic Connectivity Analysis of Biological Neural Networks at 2022-09-2
8 11:00:00
DESCRIPTION;LANGUAGE=en-US:Learning in organisms is one of their most fund
amental but intricate processes. Understanding learning is a longstanding
problem in neuroscience as well as in artificial neural networks. In this
work, we focus on studying in biological networks during motor learning th
rough the lens of the connectivity of neurons in the primary motor cortex
(M1). For this purpose, we analyze the neural activity recorded from awake
and behaving mice using two-photon calcium imaging. These imaging methods
enabled us to acquire longitudinal neuronal activity with cellular resolu
tion in many hundreds of cells in one recording session. The recordings fr
om each animal are acquired during several days, following the same neuron
s, while the animal learns a complex hand-reach task.
Our analysis is
based on a representation of the dynamic network underlying the neural ac
tivity as a sequence of graphs, where each graph describes the neuronal co
nnectivity at a certain point in time during the learning process. In this
talk, I will first present the computational methods we developed for mea
suring and quantifying the similarity between graphs in terms of their con
nectivity patterns. Then, I will show the application of these methods to
neural activity. Based on our analysis, we found that the connectivity of
the neural network in M1 smoothly converges to a steady connectivity state
as the learning process progresses, which is characterized by the formati
on of functional subsets of neurons that operate in synchrony. Moreover, w
e show that blocking the dopamine transmission from the ventral tegmental
area (VTA) to M1 disturbs the network convergence and hampers motor learni
ng. This indicates that the transition of the network requires plasticity
in the connectivity within the network.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture: 96390522948
UID:123se24012024102830
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20221002T113000
DTEND;TZID="Asia/Jerusalem":20221002T123000
DTSTAMP;TZID="Asia/Jerusalem":20221002T113000
FREEBUSY;FBTYPE=BUSY:20221002T113000/20221002T123000
SUMMARY;LANGUAGE=en-US:phd talk by Michael Amir about Multi A(ge)nt System
s on Graphs at 2022-10-02 11:30:00
DESCRIPTION;LANGUAGE=en-US:Multi-agent systems are a fascinating, multidis
ciplinary field with applications to robotics, distributed systems, biolog
y, and social dynamics. A multi-agent system is a distributed system compo
sed of several interacting, autonomous agents that cooperate to achieve so
me desired behavior. The topic of this talk is the way in which local inte
ractions between extremely simple agents can result in desirable global st
ates. We study this topic from two perspectives: the perspective of an obs
erver of the natural world, and the perspective of a designer of swarm-rob
otic systems. In the natural world, living swarms of organisms seem to eff
ortlessly and autonomously coordinate their motion. How do swarms of locus
ts converge to a single direction of motion? Why are trails of ants so str
aight and nice? As observers, our goal is to study the principles underlyi
ng these kinds of phenomena, learning what we can from Mother Nature's alg
orithms. As designers, on the other hand, our goal is to create and guaran
tee the performance of swarm-robotic systems. We seek to establish that ev
en severely myopic and computationally limited robots can be remarkably ef
fective when working together. We shall show that, with the right local al
gorithm, such robots can explore unknown environments, recover from crashe
s, split workloads, and optimize traffic systems.
Almost all mathematic
al models we work with assume the agents move in a space that is finite an
d discrete; namely, a graph environment, where spatial locations are indic
ated by vertices and the connections between them by edges. From a theoret
ical standpoint, the study of these types of multi-agent models is often a
d hoc, and relatively few general techniques are known. A main goal of our
s is to highlight a number of techniques that have proven repeatedly usefu
l in the analysis of such models, including exchangeability, coupling, pot
ential (``Lyapunov'') functions, interacting particle systems, and station
ary distributions.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture: 3603849038
UID:123se24012024102840
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20221019T103000
DTEND;TZID="Asia/Jerusalem":20221019T113000
DTSTAMP;TZID="Asia/Jerusalem":20221019T103000
FREEBUSY;FBTYPE=BUSY:20221019T103000/20221019T113000
SUMMARY;LANGUAGE=en-US:msc talk by David Vainshtein about Terraforming-Env
ironment Manipulation During Disruption Events for Lifelong Multi-Agent Pa
th Finding at 2022-10-19 10:30:00
DESCRIPTION;LANGUAGE=en-US:Lifelong Multi-Agent Pathfinding (L-MAPF) is co
ncerned with planning collision-free paths for a team of agents as they co
ntinuously handle tasks involving pick-up and delivery. When modeling auto
nomous warehouses, typical approaches for L-MAPF consider the environment
as populated with static obstacles in the form of inventory pods that the
agents must avoid. These obstacles impose narrow passageways, forcing agen
ts to resort to detours and suffer delays on account of bottlenecks. The p
roblem becomes more severe when disruptions occur: whenever an agent exper
iences a hardware failure and cannot move, or when an item drops to the wa
rehouse floor, a safety perimeter and partial blockage of the area is nece
ssary, resulting in significant detriment to performance.
In this wor
k we study a challenging case of warehouse disruptions, in which an agent
can become completely blocked but is still operational. Contemporary metho
ds can only offer to wait in place, but by manipulating the environment, t
hrough a process called Terraforming, nearby agents can be rerouted to ass
ist with mitigating disruptions. Terraforming allows agents to purposefull
y displace obstacles to create shortcuts trailblazing routes to agents who
would otherwise be considered blocked. We present a novel approach to sol
ving L-MAPF with disruptions, and propose an extension to the PBS algorith
m that incorporates Terraforming. In this manner, agents are empowered to
reason about candidate obstacles for efficient extraction, resulting in no
table improvement in throughput.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture: 94527793106
UID:123se24012024102890
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20221019T113000
DTEND;TZID="Asia/Jerusalem":20221019T123000
DTSTAMP;TZID="Asia/Jerusalem":20221019T113000
FREEBUSY;FBTYPE=BUSY:20221019T113000/20221019T123000
SUMMARY;LANGUAGE=en-US:ceClub talk by Moshe (Mickey) Gabel (York Universit
y) about ceClub: Starlight: Fast Container Provisioning on the Edge and ov
er the WAN at 2022-10-19 11:30:00
DESCRIPTION;LANGUAGE=en-US:Containers, originally designed for cloud envir
onments, are increasingly popular for provisioning workers outside the clo
ud, for example in mobile and edge computing. These settings, however, bri
ng new challenges: high latency links, limited bandwidth, and resource-con
strained workers. The result is longer provisioning times when deploying n
ew workers or updating existing ones, much of it due to network traffic.
Our analysis shows that current piecemeal approaches to reducing provi
sioning time are not always sufficient, and can even make things worse as
round-trip times grow. Rather, we find that the very same layer-based stru
cture that makes containers easy to develop and use also makes it more dif
ficult to optimize deployment. Addressing this issue thus requires rethink
ing the container deployment pipeline as a whole.
Based on our findin
gs, I will present Starlight: an accelerator for container provisioning. S
tarlight decouples provisioning from development by redesigning the contai
ner deployment protocol, filesystem, and image storage format. Our evaluat
ion using 21 popular containers shows that, on average, Starlight deploys
and starts containers 3x faster than the current industry standard impleme
ntation while incurring no runtime overhead and negligible storage overhea
d. Finally, it requires no changes to the deployed application, is backwar
ds compatible with existing workers, and uses standard container registrie
s.
Starlight is open source and available at https://github.com/mc256
/starlight.
Bio:
Moshe (Mickey) Gabel is an assistant professor in
the Department of Electrical Engineering and Computer Science at York Univ
ersity. Before joining York, he spent four years as a limited-term assista
nt professor in the Department of Computer Science at the University of To
ronto. He earned his PhD in Computer Science from the Technion – Israel In
stitute of Technology, where he also got his MSc and BSc.
Moshe’s res
earch lies in the intersection of distributed algorithms, systems, and mac
hine learning. His current research interest is edge computing, specifical
ly making geo-distributed data analysis more practical and accessible to t
ypical software developers. He has also worked extensively on machine lear
ning applications in pervasive health monitoring and in computer systems.
Moshe’s work appeared in top venues for systems and data science, includi
ng SIGMOD, NSDI, SIGKDD, VLDB, and ICML.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture: 94673013539 and Meyer 861
UID:123se24012024102900
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20221019T160000
DTEND;TZID="Asia/Jerusalem":20221019T170000
DTSTAMP;TZID="Asia/Jerusalem":20221019T160000
FREEBUSY;FBTYPE=BUSY:20221019T160000/20221019T170000
SUMMARY;LANGUAGE=en-US:phd talk by Gail Weiss about Neural Sequence Models
: A Formal Lens at 2022-10-19 16:00:00
DESCRIPTION;LANGUAGE=en-US:Neural sequence models (NSMs) - neural networks
adapted specifically for the task of processing input sequences - have em
erged as powerful tools in sequence processing, with the current most popu
lar architectures being transformers and RNN variants. But what is a train
ed network really doing? In this talk we will approach this question, star
ting from the question of what a network *can* do, and progressing to the
question of what a trained network *has* learned in practice.
Specif
ically, we will begin by discussing the mechanisms that different RNN arch
itectures can implement, and how these affect their ability to express dif
ferent formal languages. We will then move this discussion to transformers
, for which we must introduce RASP - a symbolic abstraction of their behav
iour. Finally, we will discuss how a given trained RNN can be converted to
a smaller, more interpretable, model - and how the process uncovers cases
where seemingly perfect networks have not learned their intended task!
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture: 99357013274 and Taub 601
UID:123se24012024102860
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20221020T113000
DTEND;TZID="Asia/Jerusalem":20221020T123000
DTSTAMP;TZID="Asia/Jerusalem":20221020T113000
FREEBUSY;FBTYPE=BUSY:20221020T113000/20221020T123000
SUMMARY;LANGUAGE=en-US:ceClub talk by Jongeun Lee (UNIST, Ulsan, Korea) ab
out ceClub: Challenges with Analog In-Memory Computing Arrays for Efficien
t Deep Learning Acceleration at 2022-10-20 11:30:00
DESCRIPTION;LANGUAGE=en-US:ReRAM (Resistive Random-Access Memory) crossbar
arrays have the potential to provide extremely efficient matrix-vector mu
ltiplication (MVM) operations, which are the cornerstone of many DNN (Deep
Neural Network) applications. However, there are several challenges in or
der for ReRAM crossbar arrays (RCAs) to be useful for accelerating large-s
cale DNN applications. In this talk we discuss two of those challenges. Th
e first one is the distortion in the MVM output of RCAs due to nonidealiti
es such as wire resistance (also known as IR drop) and I-V nonlinearity (i
.e., voltage-dependent conductance). While it may be very difficult to com
pletely eliminate RCA nonidealities, a fast method for system architects t
o accurately predict the output of RCAs under nonidealities would be highl
y desirable. We discuss recent methods in this direction and possible use
cases. The second challenge is the high peripheral circuit overhead of RCA
s. In particular, ADCs (Analog-Digital Converters) can account for a lion’
s share in both area and power dissipation of RCAs. I will present a quant
ization-based method, which can reduce the ADC overhead of ReRAM crossbar
arrays significantly (32x compared with ISAAC for ResNet on ImageNet datas
et) at minimal accuracy loss (of 0.24% compared with the same).
Bio:
Jongeun Lee received his B.Sc. and M.Sc. degrees in electrical engineeri
ng, and his Ph.D. in electrical engineering and computer science, all from
Seoul National University, Korea. In 2009, he joined Ulsan National Insti
tute of Science and Technology (UNIST), Ulsan, Korea, where he is now a Pr
ofessor at the Department of Electrical Engineering. Prior to joining UNIS
T, he worked as a postdoctoral research associate at Arizona State Univers
ity, Tempe, Arizona, USA, and for Samsung Electronics, Korea.
He has
published more than 80 peer-reviewed journal and conference papers and has
been on the technical program committees and organizing committees of sev
eral workshops and conferences in the areas of computer-aided design and r
econfigurable computing. His research interests are in the general area of
hardware/software co-design, and include reconfigurable architectures, de
ep learning, and computer-aided design for emerging technologies.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture: 94673013539 and Meyer 861
UID:123se24012024102910
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20221024T133000
DTEND;TZID="Asia/Jerusalem":20221024T143000
DTSTAMP;TZID="Asia/Jerusalem":20221024T133000
FREEBUSY;FBTYPE=BUSY:20221024T133000/20221024T143000
SUMMARY;LANGUAGE=en-US:cggc talk by Prof. Tatsuya Yatagawa (School of Engi
neering, The University of Tokyo) - CANCELLED! about CGGC Seminar: Deep Le
arning for Surface Geometries and Its Applications at 2022-10-24 13:30:00
DESCRIPTION;LANGUAGE=en-US:Deep learning has been one of the most common t
echniques in many industrial and research fields, and that for processing
3D geometries has also been investigated intensely in the last several yea
rs.
However, compared to the techniques for 3D volumetric images and
point sets, those for surface geometries, e.g., represented by triangular
meshes, have not been well established due to the difficulty in handling s
patial and topological features simultaneously using a neural network. In
our talk, the first part introduces several common approaches to overcomin
g the above problem to define a convolution operation on the surface geome
try, including both tessellation-aware and tessellation-agnostic approache
s. In the second part, we will introduce our recent research on denoising
polygonal meshes using graph convolutional neural networks. The proposed t
echnique does not rely on large-scale training datasets and works with onl
y noisy input mesh. The independence from training datasets allows the pro
posed method to be applied to a range of shapes uncommon in typical shape
datasets.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 401
UID:123se24012024102870
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20221026T123000
DTEND;TZID="Asia/Jerusalem":20221026T133000
DTSTAMP;TZID="Asia/Jerusalem":20221026T123000
FREEBUSY;FBTYPE=BUSY:20221026T123000/20221026T133000
SUMMARY;LANGUAGE=en-US:Theory Seminar talk by Amir Yehudayoff (Technion) a
bout Theory Seminar: Learning Dimensions at 2022-10-26 12:30:00
DESCRIPTION;LANGUAGE=en-US:We shall survey and discuss several dimensions
in the theory of machine learning.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 201
UID:123se24012024102950
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20221026T140000
DTEND;TZID="Asia/Jerusalem":20221026T150000
DTSTAMP;TZID="Asia/Jerusalem":20221026T140000
FREEBUSY;FBTYPE=BUSY:20221026T140000/20221026T150000
SUMMARY;LANGUAGE=en-US:msc talk by Reda Igbaria about Similarity-based Reg
ularization for Mitigating Artifacts at 2022-10-26 14:00:00
DESCRIPTION;LANGUAGE=en-US:Common methods for mitigating spurious correlat
ions in natural language understanding (NLU) usually operate in the output
space, encouraging a main model to behave differently from a bias model b
y down-weighing examples where the bias model is confident.
While improvi
ng out of distribution (OOD) performance, it was recently observed that th
e internal representations of the presumably debiased models are actually
more, rather than less biased.
We propose SimgReg, a new method for debi
asing internal model components via similarity-based regularization, in re
presentation space: We encourage the model to learn representations that a
re either similar to an unbiased model or different from a biased model. W
e experiment with three NLU tasks and different kinds of biases.
We find
that SimReg improves OOD performance, with little in-distribution degrada
tion. Moreover, the representations learned by SimReg are less biased than
in other methods.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture: 9533481885 and Taub 301
UID:123se24012024102930
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20221026T150000
DTEND;TZID="Asia/Jerusalem":20221026T160000
DTSTAMP;TZID="Asia/Jerusalem":20221026T150000
FREEBUSY;FBTYPE=BUSY:20221026T150000/20221026T160000
SUMMARY;LANGUAGE=en-US:msc talk by Ortal Cohen about Using Transformers to
Model Electronic Health Records of ICU and for Prediction of Blood Stream
Infection at 2022-10-26 15:00:00
DESCRIPTION;LANGUAGE=en-US:Machine learning made many recent advances in s
cience and technology, specifically in healthcare information technology.
Electronic Health Records (EHR) data store the healthcare information. EHR
data consists of many features. It is highly complicated, noisy and inclu
des many outliers and missing values. It also contains time-dependent info
rmation such as vital sign measurements, diagnosis, treatment, etc.
T
herefore, basic machine learning models have poor performance on this data
, as they do not utilize the time-series aspect of the data. We hypothesiz
e that by using the Transformer, one of the most recent advancements in ma
chine learning, we can better model EHR data, as its primary goal is to pr
ocess time-dependent data. Our model uses data in the OMOP Common data mod
el form.
Specifically, we model data related to ICU stays for predict
ing Blood Stream Infection (BSI), a critical condition with a mortality ra
te above 30%. Early prediction and antibiotic treatment of BSI are essenti
al, as it reduces mortality and morbidity significantly. We developed a Tr
ansformer-based architecture. This model was inspired by the work of Kodia
lam et al, a Transformer model which uses OMOP common data model to model
time-dependent healthcare information data.
We aim to utilize a prop
rietary dataset of about 200,000 ICU stays to identify hidden structures a
nd perform risk prediction for the BSI condition in advance.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture: 92684718772 and Taub 601
UID:123se24012024102880
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20221030T143000
DTEND;TZID="Asia/Jerusalem":20221030T153000
DTSTAMP;TZID="Asia/Jerusalem":20221030T143000
FREEBUSY;FBTYPE=BUSY:20221030T143000/20221030T153000
SUMMARY;LANGUAGE=en-US:Coding_Theory_Semina talk by Nir Weinberger (Techni
on) about Coding Theory: DNA storage: Capacity and Error Probability Bound
s at 2022-10-30 14:30:00
DESCRIPTION;LANGUAGE=en-US:We will discuss results on the capacity and err
or probability bounds of the DNA storage channel. First, we consider the c
ase in which the sequencing channel is memoryless and the coverage depth i
s constant. We will describe lower (achievability) and upper (converse) bo
unds on the capacity of the channel, as well as a lower (achievability) bo
und on the reliability function of the channel. Second, we will consider g
eneral sequencing channels, and coverage depth scaling, and focus on error
probability analysis of coding schemes which are based on coded-index. Th
e results will highlight the interaction between molecule length, coverage
depth and quality of the sequencing channel in determining the fundamenta
l limits of the DNA storage medium.
Nir Weinberger is an assistant Pr
ofessor at the The Viterbi Faculty of Electrical and Computer Engineering,
Technion – Israel Institute of Technology. Previously, from 2017 to 2018
he was a post-doctoral fellow at Tel Aviv University, and from 2018-2020 h
e was a Technion-MIT post-doctoral fellow at the Massachusetts Institute o
f Technology, Cambridge, MA, USA. He has received the B.Sc. and M.Sc. degr
ees (both summa cum laude) from Tel-Aviv University, Tel-Aviv, Israel, in
2006 and 2009, respectively, and his Ph.D. degree in 2017, from the Techni
on, Israel Institute of Technology. From 2006 to 2013 he served as an algo
rithm Engineer in the Israeli Defense Forces, working in Communications an
d Signal Processing.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 601
UID:123se24012024102940
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20221102T110000
DTEND;TZID="Asia/Jerusalem":20221102T120000
DTSTAMP;TZID="Asia/Jerusalem":20221102T110000
FREEBUSY;FBTYPE=BUSY:20221102T110000/20221102T120000
SUMMARY;LANGUAGE=en-US:phd talk by Catherine Haddad-Zaknoon about Learning
Subclasses of Junta from Membership Queries at 2022-11-02 11:00:00
DESCRIPTION;LANGUAGE=en-US:In this work, we address the problem of learnin
g subclasses of $\mathbb{JUNTA}$ under the {\it exact learning model from
membership queries} or {\it black box queries}.
First, we address the
problem of learning subclasses of decision trees from membership queries.
For adaptive non-proper learning of decision trees of depth at most $d$,
we give a randomized polynomial time algorithm that asks $\tilde O(2^{2d})
+ 2^{d}\log n$ membership queries and a deterministic polynomial time al
gorithm that asks $2^{5.83d}+2^{2d+o(d)}\log n$ queries. We develop new re
sults on the learnability of decision trees in the distribution-free model
and use those to build a random proper learning algorithm that asks $m=\t
ilde{O}(2^{2d})\log(1/\delta)+O(2^d\log n)$ queries and time $2^{O(d^2)}\l
og (1/\delta) +O(mn)$, and a deterministic proper learning algorithm for
learning the class $\DT_d^s$ with $m=2^{2d+o(d)}\log n$ queries and runnin
g time of $s^{O(d)}+O(mn)$.
Second, we consider the problem of group
testing problem with non-adaptive randomized algorithms. Several models h
ave been discussed to determine how to randomly choose the tests. For a mo
del ${\cal M}$, let $m_{\cal M}(n,d)$ be the minimum number of tests requi
red to detect at most $d$ defectives within $n$ items, with success probab
ility at least $1-\delta$, for some constant $\delta$. We study the measur
es
$$c_{\cal M}(d)=\lim_{n\to \infty} \frac{m_{\cal M}(n,d)}{\ln n}
\m
box{\ and \ } c_{\cal M}=\lim_{d\to \infty} \frac{c_{\cal M}(d)}{d}.$$
Our analyses yields tight bounds for $c_{\cal M}(d)$ and $c_{\cal M}$ f
or all the known models~${\cal M}$.
Third, we examine the problem of
estimating the number of defective items $d$ up to a multiplicative facto
r $\Delta>1$, using deterministic group testing when some upper bound $D\g
eq d$ is known. We bring lower and upper bounds on the number of tests req
uired in both the adaptive and the non-adaptive deterministic settings. Fo
r the adaptive deterministic settings, we prove a lower bound of $\Omega \
left((D/\Delta^2)\log (n/D) \right )$ tests. Moreover, we give a polynomi
al time adaptive algorithm that shows that our bound is tight up to a smal
l additive term. For non-adaptive algorithms, an upper bound of $O((D/\Del
ta^2)$ $(\log (n/D)+\log \Delta) )$ is achieved by means of non-constructi
ve proof. This result matches the lower bound up to a small additive term.
In addition, we use existing polynomial time constructible \emph{expander
regular bipartite graphs}, \emph{extractors} and \emph{condensers} to con
struct two polynomial time algorithms. The first makes $O((D^{1+o(1)}/\Del
ta^2)\cdot \log n)$ tests, and the second makes $(D/\Delta^2)\cdot Quazipo
ly$ $(\log n)$ tests. This is the first explicit construction with an almo
st optimal test complexity.
Moreover, we study the problem of determi
ning exactly the number of defective items in an adaptive group testing by
using a minimum number of tests. We improve the existing algorithm and pr
ove a lower bound that shows that the number of tests in our algorithm is
optimal up to small additive terms. In particular, we prove that any rando
mized Monte Carlo algorithm for this task must ask at least
$$ \left(
1-\frac{\log d+\log(1/\delta)+1}{\log n+\log(1/\delta)}\right)d\log (1/2\d
elta)$$
queries. Moreover, we prove an upper bound of $(1+o(1))d\log(d/\
delta)$ for the same settings. This bound is optimal up to the additive te
rm $(1+o(1))d\log d$.
Finally, we consider some applicable aspects of
group testing. We propose a heuristic random method to construct the test
design. Using the suggested design, along with group testing – compressiv
e sensing decoding method, our experiments imply that, for some values of
the pool size, a reduction of up to 10 fold in the tests number can be ach
ieved. We discuss the applicability of this process in accelerating medica
l tests required for COVID-19 PCR testing.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture: 4966839889
UID:123se24012024102850
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20221102T123000
DTEND;TZID="Asia/Jerusalem":20221102T133000
DTSTAMP;TZID="Asia/Jerusalem":20221102T123000
FREEBUSY;FBTYPE=BUSY:20221102T123000/20221102T133000
SUMMARY;LANGUAGE=en-US:Theory Seminar talk by Yaroslav Alekseev (St. Peter
sburg university) about Theory Seminar: The power of the Binary Value Prin
ciplein Proof Complexity at 2022-11-02 12:30:00
DESCRIPTION;LANGUAGE=en-US:The (extended) Binary Value Principle (eBVP: $k
+ x_0 + 2x_1 + … + 2^n x_n $ for $k>0$ and $x^2_i=x_i$) has received a lo
t of attention recently: several lower bounds have been proved for it (Ale
kseev et al 2020, Alekseev 2021, Part and Tzameret 2021),
and a polynomi
al simulation of a strong semialgebraic proof system in IPS+eBVP has been
shown (Alekseev et al 2020).
In this talk, we consider Polynomial Cal
culus with the algebraic version of Tseitin’s extension rule. We show that
in this context eBVP still allows to simulate similar semialgebraic syste
ms. We also prove that it allows to simulate the Square Root Rule (Grigori
ev and Hirsch 2003),
which is absolutely unclear in the context of ordin
ary Polynomial Calculus.
On the other hand, we demonstrate that eBVP
probably does not help in proving exponential lower bounds for Boolean tau
tologies: we show that an Extended Polynomial Calculus derivation of any s
uch tautology from eBVP must be of exponential size.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 201
UID:123se24012024102980
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20221106T143000
DTEND;TZID="Asia/Jerusalem":20221106T153000
DTSTAMP;TZID="Asia/Jerusalem":20221106T143000
FREEBUSY;FBTYPE=BUSY:20221106T143000/20221106T153000
SUMMARY;LANGUAGE=en-US:Coding_Theory_Semina talk by Orian Leitersdorf (Tec
hnion) about Coding Theory: Codes for Constrained Periodicity at 2022-11-0
6 14:30:00
DESCRIPTION;LANGUAGE=en-US:Reliability is an inherent challenge for the em
erging nonvolatile technology of racetrack memories, and there exists a fu
ndamental relationship between codes designed for racetrack memories and c
odes with constrained periodicity. Previous works have sought to construct
codes that avoid periodicity in windows, yet have either only provided ex
istence proofs or required high redundancy. This paper provides the first
constructions for avoiding periodicity that are both efficient (average-li
near time) and with low redundancy (near the lower bound). The proposed al
gorithms are based on iteratively repairing windows which contain periodic
ity until all the windows are valid. Intuitively, such algorithms should n
ot converge as there is no monotonic progression; yet, we prove convergenc
e with average-linear time complexity by exploiting subtle properties of t
he encoder. Overall, we both provide constructions that avoid periodicity
in all windows, and we also study the cardinality of such constraints.
Orian Leitersdorf is currently studying towards the B.Sc, M.Sc., and Ph.
D. degrees at the Technion, Haifa, Israel. He is a scholar at both the Tec
hnion Excellence Program and the Technion Lapidim CS Excellence Program, a
nd is also a recipient of the Gutwirth Excellence Scholarship. His current
research aims to advance digital processing-in-memory (PIM) towards funda
mental applications (e.g., matrix operations, graph algorithms, cryptograp
hy) while also addressing challenges such as reliability. Further, his res
earch interests also include several topics in theoretical computer scienc
e, including coding theory.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 601
UID:123se24012024103010
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20221107T110000
DTEND;TZID="Asia/Jerusalem":20221107T120000
DTSTAMP;TZID="Asia/Jerusalem":20221107T110000
FREEBUSY;FBTYPE=BUSY:20221107T110000/20221107T120000
SUMMARY;LANGUAGE=en-US:msc talk by Itay Eilat about Strategic Classificati
on with Inter-Dependent Strategic Responses at 2022-11-07 11:00:00
DESCRIPTION;LANGUAGE=en-US:Strategic classification studies learning in se
ttings where agents can modify their features to obtain favourable outcome
. Most current works focus on simple decision rules that trigger independe
nt agent responses. Here we examine the implications of learning with more
elaborate models that break the independence assumption. We present two w
orks, each studying different (but related) models and tasks, and in which
dependencies are introduces through space (using graphs) and time. Our fi
rst work considers learning with graph neural networks (GNNs) – neural arc
hitectures which make use of social relations between users to improve cla
ssification. We show how relying on the graph inadvertently introduces int
er-user dependencies that significantly affect predictive outcomes. Our se
cond work focuses on recommendation, and studies how learning personalized
item scores incentivizes exposure-maximizing content creators to update t
heir items. We argue that this incentive structure can essentially elimina
te diversity in recommendations, but at the same time, can be used to prom
ote intrinsic diversity over time – a process in which the underlying user
-item graph is key. In both works, we use analysis and simulative experime
nts to show how strategic responses can either work against the system – o
r in its favor, and propose differential frameworks for learning in ways t
hat promote both system and societal interests.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture: 6048862164 and Taub 601
UID:123se24012024102970
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20221108T183000
DTEND;TZID="Asia/Jerusalem":20221108T203000
DTSTAMP;TZID="Asia/Jerusalem":20221108T183000
FREEBUSY;FBTYPE=BUSY:20221108T183000/20221108T203000
SUMMARY;LANGUAGE=en-US:CSpecial Event about The Real Life behind Entrepren
eurship at 2022-11-08 18:30:00
DESCRIPTION;LANGUAGE=en-US:You are invited to a lecture by Iris Shor about
the real life behind entrepreneurship at Oribi, who will tell about her p
ath as an entrepreneur and CEO and about life as it is for entrepreneurs,
on Tuesday, November 8, 2022, at 18:30 on Taub Terrace.
The lecture i
s open to all CS students.
Please pre-register
Pictures from the event
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Room 337 Taub Bld.
UID:123se24012024103000
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20221109T123000
DTEND;TZID="Asia/Jerusalem":20221109T133000
DTSTAMP;TZID="Asia/Jerusalem":20221109T123000
FREEBUSY;FBTYPE=BUSY:20221109T123000/20221109T133000
SUMMARY;LANGUAGE=en-US:Theory Seminar talk by Kasper Green Larsen (Aarhus
university) about Theory Seminar: Optimal Weak to Strong Learning at 2022-
11-09 12:30:00
DESCRIPTION;LANGUAGE=en-US:The classic algorithm AdaBoost allows to conver
t a weak learner, that is an algorithm that produces a hypothesis which is
slightly better than chance, into a strong learner, achieving arbitrarily
high accuracy when given enough training data. We present a new algorithm
that constructs a strong learner from a weak learner but uses less traini
ng data than AdaBoost and all other weak to strong learners to achieve the
same generalization bounds. A sample complexity lower bound shows that ou
r new algorithm uses the minimum possible amount of training data and is t
hus optimal. Hence, this work settles the sample complexity of the classic
problem of constructing a strong learner from a weak learner.
This w
ork was accepted at NeurIPS’22 and is joint work with Martin Ritzert from
Aarhus University.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 201
UID:123se24012024103020
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20221109T123000
DTEND;TZID="Asia/Jerusalem":20221109T143000
DTSTAMP;TZID="Asia/Jerusalem":20221109T123000
FREEBUSY;FBTYPE=BUSY:20221109T123000/20221109T143000
SUMMARY;LANGUAGE=en-US:CSpecial Event about Recruitment Day by DELL Techn
ologies at 2022-11-09 12:30:00
DESCRIPTION;LANGUAGE=en-US:You are invited to a Recruitment Day by DELL Te
chnologies to meet engineers and recruit teams who will present programs
for graduates and students and their open jobs, on Wednesday, November 9,
20022, 12:30-14:30 at CS Lobby.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:CS Taub Lobby
UID:123se24012024103050
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20221113T133000
DTEND;TZID="Asia/Jerusalem":20221113T143000
DTSTAMP;TZID="Asia/Jerusalem":20221113T133000
FREEBUSY;FBTYPE=BUSY:20221113T133000/20221113T143000
SUMMARY;LANGUAGE=en-US:cggc talk by Florine Hartwig (University of Bonn) a
bout CGGC Seminar: Shape Matching of Discrete Shells at 2022-11-13 13:30:0
0
DESCRIPTION;LANGUAGE=en-US:Finding correspondences between shapes is a cen
tral task in geometry processing with many applications such as texture or
deformation transfer and shape interpolation. We focus on developing a me
thod to find correspondences between non-isometric geometric shapes. Our m
ethod follows the functional map approach. However, unlike existing classi
cal functional map approaches our method is able to match extrinsic surfac
e features by design. To achieve this we consider eigenfunctions of the He
ssian of an elastic thin shell energy to construct a new reduced basis for
the function spaces occurring in this context.
Florine Hartwig recently
obtained her Master’s degree in Mathematics from the University of Bonn f
ocusing on Numerical Analysis. She is starting her Ph.D. in the working gr
oup of Prof. Dr. Martin Rumpf at the University of Bonn working on extendi
ng the theory of Riemannian Shape Spaces and problems in geometry processi
ng.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Room 337 Taub Bld.
UID:123se24012024102920
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20221114T173000
DTEND;TZID="Asia/Jerusalem":20221114T193000
DTSTAMP;TZID="Asia/Jerusalem":20221114T173000
FREEBUSY;FBTYPE=BUSY:20221114T173000/20221114T193000
SUMMARY;LANGUAGE=en-US:CSpecial Event about All You Wanted to Know about
a Master's Degree at 2022-11-14 17:30:00
DESCRIPTION;LANGUAGE=en-US:You are invited to a question and answer sessio
n about all that interests you in your master's degree, with a panel of CS
graduate students, who will also give you tips and share with you their e
xperience and challenges, with the participation of CS Ph.D. students: Ran
a Shahout, Hadas Orgad, Eden Seig and Omar Sabary, On Monday, November 14,
2022, at 17:30 at the Graduates Club (at the end of the corridor on the
2nd floor).
The meeting is intended for master's degree students and
students nearing the completion of their bachelor's degree who are interes
ted in further studies for a master's degree.
Please register in advance.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Graduates Club (Taub Build., Floor2)
UID:123se24012024103080
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20221116T123000
DTEND;TZID="Asia/Jerusalem":20221116T133000
DTSTAMP;TZID="Asia/Jerusalem":20221116T123000
FREEBUSY;FBTYPE=BUSY:20221116T123000/20221116T133000
SUMMARY;LANGUAGE=en-US:Theory Seminar talk by Lianna Hambardzumyan (The He
brew University of Jerusalem) about Theory Seminar: Dimension-free Relatio
ns in Communication Complexity at 2022-11-16 12:30:00
DESCRIPTION;LANGUAGE=en-US:In this talk we will discuss dimension-free rel
ations between basic communication and query complexity measures and vario
us matrix norms. Dimension-free relations are inequalities that bound a pa
rameter as a function of another parameter without dependency on the numbe
r of input bits. This is in contrast to the more common framework in commu
nication complexity where polylogarithmic dependencies are tolerated. Dime
nsion-free bounds are closely related to structural results, where one see
ks to describe the structure of Boolean matrices and functions that have “
low complexity”. We prove such bounds for several communication and query
complexity measures as well as various matrix and operator norms. We also
propose several conjectures, and establish connections to graph theory, op
erator theory, and harmonic analysis.
The talk is based on joint work wi
th Hamed Hatami and Pooya Hatami.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 201
UID:123se24012024103090
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20221116T130000
DTEND;TZID="Asia/Jerusalem":20221116T140000
DTSTAMP;TZID="Asia/Jerusalem":20221116T130000
FREEBUSY;FBTYPE=BUSY:20221116T130000/20221116T140000
SUMMARY;LANGUAGE=en-US:msc talk by Liam B. Kimel about Thermodynamic Model
s for the Homeostasis And Development of the Epidermis and Spherical Organ
oids at 2022-11-16 13:00:00
DESCRIPTION;LANGUAGE=en-US:Epithelial tissues formed of layered cell surfa
ces are prevalent in multiple tissues and play an essential role in key bi
ological processes such as development, organ homeostasis and cancer.
In this work we study the mechanical aspects of layered multi-cell syst
ems using pseudo-thermodynamic models.
First, by use of stochastic s
imulations we were able to characterize the proliferation properties of a
new type of stem cell in the mouse interfollicular epidermis.
Furthe
r, we used both deterministic and stochastic computer simulations to inves
tigate the interplay and between the biological and mechanical properties
of concentric, multilayered spherical cell systems.
We found that lo
calized active contractility in the outer cell layers is critical in maint
aining system homeostasis, and further controls dynamic properties such as
the system growth rate and fluctuations about the steady-state.
Our
findings go towards gaining a mechanistic understanding of multilayered e
pithelial systems and suggest strategies by which they may be modulated.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture: 93465319106
UID:123se24012024102960
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20221120T133000
DTEND;TZID="Asia/Jerusalem":20221120T143000
DTSTAMP;TZID="Asia/Jerusalem":20221120T133000
FREEBUSY;FBTYPE=BUSY:20221120T133000/20221120T143000
SUMMARY;LANGUAGE=en-US:cggc talk by Rephael Wenger (The Ohio State Univers
ity) about CGGC Seminar: Isosurfaces: Fast Generation of Large Geometric M
odels from Scalar Data at 2022-11-20 13:30:00
DESCRIPTION;LANGUAGE=en-US:Isosurfaces are surface meshes representing ”ob
ject” boundaries in scalar field data, such as MRI or CT data. More precis
ely, isosurfaces are surface meshes representing level sets {f−1(σ) : σ ∈
R}, constructed from a sampling of a function f : R → R3. Isosurface const
ruction, particularly the classical Marching Cubes algorithm (1988), is a
standard tool in scientific visualization and geometric modeling, We will
discuss Marching Cubes and an alternative, Dual Contouring, isosurface con
struction in 4D, extending these algorithms to volume meshing (interval vo
lumes), representing and constructing sharp features using these algorithm
s, and modifying these algorithms to construct good (better) quality trian
gle or quadrilateral meshes.
Rephael Wenger is a professor in the com
puter science and engineering department of The Ohio State University wher
e he works on geometric modeling, mesh generation, geometric algorithms an
d scientific visualization. The focus of his recent research has been on t
he fast generation of large geometric models (isosourfaces) from scalar da
ta such as industrial and medical CT data, MRI data, time varying (4D) sca
lar data and computational fluid dynamics simulation data. He wrote a book
and has authored numerous papers on isosurface generation, including pape
rs on isosurface generation in 4D, generating 3D meshes bounded by isosurf
aces (interval volumes), generating isosurfaces with guarantees on mesh qu
ality, fractal dimensions of isosurfaces, and generating isosurfaces with
sharp features.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 301
UID:123se24012024103070
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20221121T133000
DTEND;TZID="Asia/Jerusalem":20221121T143000
DTSTAMP;TZID="Asia/Jerusalem":20221121T133000
FREEBUSY;FBTYPE=BUSY:20221121T133000/20221121T143000
SUMMARY;LANGUAGE=en-US:msc talk by Saar Huberman about Learning to Efficie
ntly Compute Accurate Geodesic Distances on Surfaces at 2022-11-21 13:30:0
0
DESCRIPTION;LANGUAGE=en-US:A high order accurate deep learning method for
computing geodesic distances on surfaces is introduced. We consider two ma
in components for computing distances on surfaces; A numerical solver that
locally approximates the distance function and an efficient causal orderi
ng scheme by which surface points are updated. The proposed method exploit
s a dynamic programming principle which lends itself to a scheme with quas
i-linear computational complexity. The quality of the distance approximati
on is determined by the local solver and is the main focus of our research
. A common approach to compute distances on continuous surfaces is by cons
idering a discretized polygonal mesh approximating the surface, and estima
ting distances on the polygon. With such an approximation, the exact geode
sic distances restricted to the polygon are at most second order accurate
with respect to the distances on the corresponding continuous surface. Her
e, by order of accuracy we refer to the rate of convergence as a function
of the average distance between sampled points. To improve the accuracy, w
e consider a neural network based local solver which implicitly approximat
es the structure of the continuous surface. The proposed solver circumvent
s the polyhedral representation, by directly consuming sampled mesh vertic
es for approximation of distances on the sampled continuous surfaces. We s
upply numerical evidence that the proposed learned update scheme, with app
ropriate local numerical support, provides better accuracy compared to the
best possible polyhedral approximations and previous learning based metho
ds. We introduce a trained solver which is third order accurate, with quas
i-linear complexity in the number of sampled points.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture: 9811231512
UID:123se24012024102990
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20221121T183000
DTEND;TZID="Asia/Jerusalem":20221121T203000
DTSTAMP;TZID="Asia/Jerusalem":20221121T183000
FREEBUSY;FBTYPE=BUSY:20221121T183000/20221121T203000
SUMMARY;LANGUAGE=en-US:CSpecial Event about Pikoya Lecture: Building the
Monstera Games Platform at 2022-11-21 18:30:00
DESCRIPTION;LANGUAGE=en-US:You are invited to the lecture: Building the Mo
nstera Games Platform by David Yanai, CTO at Pikoya and CS magister graduate, about the world and the gaming industry used by bil
lions of users worldwide, and about the implementation aspects, the archit
ecture and the many challenges that Pikoya faced in building the platform.
The lecture will take place on Monday, November 21, 2022, at 18:30 p
m in Taub 337.
Please pre-register.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Room 337 Taub Bld.
UID:123se24012024103130
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20221122T143000
DTEND;TZID="Asia/Jerusalem":20221122T153000
DTSTAMP;TZID="Asia/Jerusalem":20221122T143000
FREEBUSY;FBTYPE=BUSY:20221122T143000/20221122T153000
SUMMARY;LANGUAGE=en-US:colloq talk by David Wajc (Google Research → Techni
on) about CS Colloquia: Dynamic Matching with Better-than-2 Approx in Poly
log Update Time at 2022-11-22 14:30:00
DESCRIPTION;LANGUAGE=en-US:We present dynamic algorithms with polylogarith
mic update time for estimating the size of the maximum matching of a graph
undergoing edge insertions and deletions with approximation ratio strictl
y better than 2. This answers in the affirmative the value version of a ma
jor open question, repeatedly asked in the dynamic graph algorithms litera
ture.
Based on an upcoming SODA 2023 best paper, joint with Sayan Bh
attacharya, Peter Kiss and Thatchaphol Saranurak.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Room 337 Taub Bld.
UID:123se24012024103120
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20221123T113000
DTEND;TZID="Asia/Jerusalem":20221123T123000
DTSTAMP;TZID="Asia/Jerusalem":20221123T113000
FREEBUSY;FBTYPE=BUSY:20221123T113000/20221123T123000
SUMMARY;LANGUAGE=en-US:msc talk by Raïssa Nataf about Null Messages, Infor
mation and Coordination at 2022-11-23 11:30:00
DESCRIPTION;LANGUAGE=en-US:This work investigates the transfer of informat
ion in fault-prone synchronous systems using null messages.
The notion
of an {\em $f$-resilient message block} is defined to capture the fundame
ntal communication pattern for knowledge transfer. This pattern may involv
e null messages in addition to explicit messages, and hence, it provides a
fault-tolerant extension of the classic notion of a message-chain.
Ba
sed on the above, we provide tight necessary and sufficient characterizati
ons of the generalized communication patterns, including actual messages
and null messages, that can serve to solve the distributed tasks of (nice-
run) Signalling and Ordered Response.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture: 96118036561 and Taub 401
UID:123se24012024103100
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20221123T123000
DTEND;TZID="Asia/Jerusalem":20221123T143000
DTSTAMP;TZID="Asia/Jerusalem":20221123T123000
FREEBUSY;FBTYPE=BUSY:20221123T123000/20221123T143000
SUMMARY;LANGUAGE=en-US:CSpecial Event about Open Day for Graduate Studies
at CS at 2022-11-23 12:30:00
DESCRIPTION;LANGUAGE=en-US:Hello to outstanding undergraduate students at
CS!
CS invites undergraduate outstanding students to continue the tr
adition of excellence and participate in a meeting on postgraduate studies
at the faculty.
The meeting will present the multitude of options a
nd opportunities offered to advanced degrees students, and will be held on
Wednesday, November 23, 2022, 12:30-14::00 Taub 337, 3rd floor, Taub Comp
uter Science Building. CS Acting Dean and the Vice Dean for graduate studi
es will participate in the meeting and will be happy to answer your questi
ons.
In the program:
12:30-12:45 - Gathering
12:45-12:50 - Pro
f. Danny Raz, Acting Dean of the Faculty of Computer Science: Opening Rema
rks
12:50-13:10 - Prof. Gil Barequet, Vice Dean for Graduate Studies: "W
hat is a Master's Degree and what is a Doctorate?"
13:10-13:30 - Dr.
Yonatan Belinkov: "Advanced Degree on the Way to Natural Language Process
ing"
13:30-13:40 - Ms. Hadas Orgad (CS Ph.D. student): "Research and Lif
e at CS"
13:40- - Questions and answers
Those interested
in attending the meeting are asked to register in advance<
/a>.
For questions: Limor Gindin, phone 04-8294226, email: limorg@cs.technion.ac.il
More de
tails about graduate studies at the faculty: graduate.cs.technion.ac.il
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Room 337 Taub Bld.
UID:123se24012024103040
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20221123T123000
DTEND;TZID="Asia/Jerusalem":20221123T133000
DTSTAMP;TZID="Asia/Jerusalem":20221123T123000
FREEBUSY;FBTYPE=BUSY:20221123T123000/20221123T133000
SUMMARY;LANGUAGE=en-US:Theory Seminar talk by Amir Abboud (Weizmann Instit
ute of Science) about Theory Seminar: APMF < APSP? Gomory-Hu Tree in Subcu
bic Time at 2022-11-23 12:30:00
DESCRIPTION;LANGUAGE=en-US:The All-Pairs Max-Flow problem (APMF) asks to c
ompute the maximum flow (or equivalently, the minimum cut) between all pai
rs of nodes in a graph. The naive solution of making n^2 calls to a (singl
e-pair) max-flow algorithm was beaten in 1961 by a remarkable algorithm of
Gomory and Hu that only makes n-1 calls. Within the same time bound, thei
r algorithm also produces a cut-equivalent tree (a.k.a. GH-Tree) that pres
erves all pairwise minimum cuts exactly. This gives a cubic upper bound fo
r APMF assuming that single-pair max-flow can be solved optimally and the
only improvements since 1961 have been on getting us closer to this assump
tion; new algorithms that break the cubic barrier were only known for spec
ial graph classes or with approximations.
The All-Pairs Shortest-Path
s problem (APSP) is similar, but asks to compute the distance rather than
the connectivity between all pairs of nodes. Its time complexity also appe
ars similar, with classical cubic time algorithms that have only been brok
en in special cases or with approximations. Meanwhile, in the past 10 year
s, the conjecture that APSP requires cubic time has played a central role
in fine-grained complexity, leading to cubic conditional lower bounds for
many other fundamental problems that appear even easier than APMF. However
, a formal reduction from APSP to APMF has remained elusive.
This tal
k will survey recent progress (based on joint works with Robert Krauthgame
r and Ohad Trabelsi), starting with partially successful attempts at reduc
ing APSP to APMF, going through algorithmic progress on APMF in limited se
ttings, and leading up to a very recent paper (also with Jason Li, Debmaly
a Panigrahi, and Thatchaphol Saranurak) where we break the 60-year old cub
ic barrier for APMF; suggesting a separation between APMF and APSP.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 201
UID:123se24012024103150
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20221123T183000
DTEND;TZID="Asia/Jerusalem":20221123T203000
DTSTAMP;TZID="Asia/Jerusalem":20221123T183000
FREEBUSY;FBTYPE=BUSY:20221123T183000/20221123T203000
SUMMARY;LANGUAGE=en-US:CSpecial Event about "Research on the Bar" Evening
- TED Lectures at 2022-11-23 18:30:00
DESCRIPTION;LANGUAGE=en-US:You are invited to the "Research on the Ba
r" evening - TED lectures and a meeting with three faculty members an
d their research groups on Wednesday, November 2
3, 2022 at 18:30 pm in Taub 337:
Dr. Sarah
Kern: Does it pay for my robot to be nice?
Dr. Yaniv Romano: Polygraph f
or learning systems: even machines make mistakes sometimes
Dr. Nir Rosen
feld: Systems learn in a human environment, or: Who needs a phone cradle?
Please register i
n advance
Pictures from the event
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:CS Taub Build. Auditorium 1
UID:123se24012024103140
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20221127T154500
DTEND;TZID="Asia/Jerusalem":20221127T164500
DTSTAMP;TZID="Asia/Jerusalem":20221127T154500
FREEBUSY;FBTYPE=BUSY:20221127T154500/20221127T164500
SUMMARY;LANGUAGE=en-US:Coding_Theory_Semina talk by Daniella Bar-Lev (CS,
Technion) about Coding Theory: Generalized Unique Reconstruction from Subs
trings at 2022-11-27 15:45:00
DESCRIPTION;LANGUAGE=en-US:New families of reconstruction codes motivated
by DNA data storage and sequencing applications will be discussed. In such
applications, DNA strands are sequenced by reading some subset of their s
ubstrings. The discussion will start with the extreme case of the torn-pap
er channel in which substrings are read with no overlap. Our model extends
the previously researched probabilistic setting to the worst-case. We wil
l construct asymptotically optimal codes, with efficient encoding and deco
ding algorithms, for any parameters of the torn-paper channel, for which a
non vanishing asymptotic rate is possible. In contrast to no overlaps bet
ween the read substrings, we consider the other extreme case in which all
substrings of a few pre-defined lengths are read. Previously researched mo
dels are extended by studying the setup where multiple strings are reconst
ructed together. Two constructions of such multi-strand reconstruction cod
es, with rates which approach 1, will be presented. Finally, an extension
for the model that considers the setup, in which consecutive substrings ar
e read with some given minimum overlap, will be discussed. An upper bound
on the attainable rates of codes that guarantee unique reconstruction will
be given. Efficient constructions of codes that asymptotically meet this
bound will be presented.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 601
UID:123se24012024103110
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20221130T123000
DTEND;TZID="Asia/Jerusalem":20221130T133000
DTSTAMP;TZID="Asia/Jerusalem":20221130T123000
FREEBUSY;FBTYPE=BUSY:20221130T123000/20221130T133000
SUMMARY;LANGUAGE=en-US:Theory Seminar talk by Irit Dinur (Weizmann Institu
te of Science) & Gil Cohen (Tel-Aviv university) about Theory Seminar: Ran
dom Walkson Rotating Expanders at 2022-11-30 12:30:00
DESCRIPTION;LANGUAGE=en-US:Random walks on expanders are extremely useful
in TOC. Unfortunately though, they have an inherent cost. E.g., the spectr
al expansion of a Ramanujan graph deteriorates exponentially with the leng
th of the walk (when compared to a Ramanujan graph of the same degree). In
this talk, we will see how this exponential cost can be reduced to linear
by applying a permutation after each random step. These permutations are
tailor-made to the graph at hand, requiring no randomness to generate. Our
proof is established using the powerful framework of finite free probabil
ity and interlacing families that was introduced, around ten years ago, by
Marcus, Spielman, and Srivastava in their seminal sequence of works on th
e existence of bipartite Ramanujan graphs of every size and every degree,
and in their solution to the Kadison-Singer problem.
Joint work with
Gal Maor.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 201
UID:123se24012024103170
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20221207T113000
DTEND;TZID="Asia/Jerusalem":20221207T123000
DTSTAMP;TZID="Asia/Jerusalem":20221207T113000
FREEBUSY;FBTYPE=BUSY:20221207T113000/20221207T123000
SUMMARY;LANGUAGE=en-US:phd talk by Rana Shahout about Sketching Streaming
Data at 2022-12-07 11:30:00
DESCRIPTION;LANGUAGE=en-US:Stream monitoring is fundamental in many data s
tream applications, such as financial data trackers, security, anomaly det
ection, and load balancing. To cope with high-speed data streams, these ap
plications require algorithms that are both time and space efficient to co
pe with high-speed data streams. Space efficiency is needed due to the mem
ory hierarchy structure, to enable cache residency and to avoid page swapp
ing. Even if the potential computing cost is low, this residency is critic
al for good performance (e.g., constant time algorithms may be inefficient
if they access the DRAM for each element). To that end, stream processing
algorithms often build compact approximate sketches (synopses) of the inp
ut streams.
This work improves the speed and space requirements for s
treaming problems.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture: 99794111202 and Taub 301
UID:123se24012024103190
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20221207T113000
DTEND;TZID="Asia/Jerusalem":20221207T123000
DTSTAMP;TZID="Asia/Jerusalem":20221207T113000
FREEBUSY;FBTYPE=BUSY:20221207T113000/20221207T123000
SUMMARY;LANGUAGE=en-US:ceClub talk by Yoav Etsion (CS & EE, Technion) abou
t ceClub: Accelerating Big Data Analytics with the Speedata Analytics Proc
essing Unit (APU at 2022-12-07 11:30:00
DESCRIPTION;LANGUAGE=en-US:At a time when Moore’s Law is reaching its end,
the volume of data created, curated, and consumed worldwide grows exponen
tially. As a result, general-purpose processors are struggling to keep up
with the throughput demand of big data analytics workloads.
In this t
alk I will present Speedata and the Analytics Processing Unit (APU), the f
irst hardware accelerator for big data analytics. Speedata’s APU accelerat
es existing software frameworks (e.g., Apache Spark, Apache Presto) and im
proves performance/cost by orders-of-magnitude while maintaining users’ ex
isting code base.
I will present how the APU’s programmable architec
ture supports high-bandwidth processing of complex analytics queries. And
after the unavoidable technical part, I will try to share some of my insig
hts on the crazy journey of starting a semiconductor company.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Room 861, EE Meyer Building
UID:123se24012024103220
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20221207T123000
DTEND;TZID="Asia/Jerusalem":20221207T143000
DTSTAMP;TZID="Asia/Jerusalem":20221207T123000
FREEBUSY;FBTYPE=BUSY:20221207T123000/20221207T143000
SUMMARY;LANGUAGE=en-US:CSpecial Event about Recruitment Day by Biosense W
ebster at 2022-12-07 12:30:00
DESCRIPTION;LANGUAGE=en-US:Engineers and recruitment teams from Biosense W
ebster will visit CS to offer employment options and open positions, and t
o lecture on the connection between medicine and computer science.
Wednesday, December 7, 2022, in the lobby and Taub 3 at CS Taub Building:
12:30 in Taub Lobby - meeting with engineers and the recruitment teams
13:30 in Taub 3 - Lecture on Geometrical Challenges in Treating Arrythmia
by Dr. Fadi Matzareva, CS graduate and software engineer in the 3D field
at Business Webster.
Please reg
ister in advance
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:CS Taub Lobby & Taub 3
UID:123se24012024103250
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20221207T133000
DTEND;TZID="Asia/Jerusalem":20221207T143000
DTSTAMP;TZID="Asia/Jerusalem":20221207T133000
FREEBUSY;FBTYPE=BUSY:20221207T133000/20221207T143000
SUMMARY;LANGUAGE=en-US:Theory Seminar talk by Moran Feldman (Haifa univers
ity) about Theory Seminar: Streaming Algorithms for Submodular Maximizatio
with a Cardinality Constraint at 2022-12-07 13:30:00
DESCRIPTION;LANGUAGE=en-US:Motivated by machine learning applications, muc
h research over the last decade was devoted to solving submodular maximiza
tion problems under Big Data computational models. Perhaps the most basic
such problem is the problem of maximizing a submodular function subject to
a cardinality constraint. A recent series of papers has studied this prob
lem in the data stream model, and in particular, fully determined the appr
oximation ratio that can be obtained for it by (semi-)-streaming algorithm
s both when the objective function is monotone and non-monotone. In this t
alk we will survey the main ideas behind these tight results.
Based o
n joint works with Naor Alaluf, Alina Ene, Huy Nguyen, Ashkan Norouzi-Fard
, Andrew Suh, Ola Svensson and Rico Zenklusen.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 201
UID:123se24012024103240
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20221207T183000
DTEND;TZID="Asia/Jerusalem":20221207T203000
DTSTAMP;TZID="Asia/Jerusalem":20221207T183000
FREEBUSY;FBTYPE=BUSY:20221207T183000/20221207T203000
SUMMARY;LANGUAGE=en-US:CSpecial Event about Personal Branding Workshop: H
ow to do it Right? at 2022-12-07 18:30:00
DESCRIPTION;LANGUAGE=en-US:You are invited to a workshop by Raviv Gortenstein (Director of People Brand & Experience a
t Riskified) who will present practical tools fo
r building a personal brand and how it can help in building a career, expa
nding the networkng, finding opportunities and positioning expertise in th
e personal field.
The workshop will be held on Wednesday, December 7
, 2022, 18:30, in Taub 337.
Please register in advance.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Room 337 Taub Bld.
UID:123se24012024103230
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20221211T143000
DTEND;TZID="Asia/Jerusalem":20221211T153000
DTSTAMP;TZID="Asia/Jerusalem":20221211T143000
FREEBUSY;FBTYPE=BUSY:20221211T143000/20221211T153000
SUMMARY;LANGUAGE=en-US:Coding_Theory_Semina talk by Roni Con (Tel-Aviv Uni
versity) about Coding Theory: Reed-Solomon Codes Against Adversarial Inser
tions and Deletions at 2022-12-11 14:30:00
DESCRIPTION;LANGUAGE=en-US:The task of constructing codes against adversar
ial insertions and deletions (insdel) has recently received much attention
.
In this work, we study the performance of Reed-Solomon codes agains
t insdel errors. We prove that there are Reed-Solomon codes that achieve t
he half-Singleton bound. In other words, there are optimal Reed-Solomon co
des also against insdel errors. We also give a set of evaluation points th
at define a Reed-Solomon code that achieves this bound. As the field size
that we get grows very fast, our construction runs in polynomial time only
for very small values of $k$, the dimension of the code.
We also exp
licitly construct two-dimensional Reed-Solomon codes over a field of size
$O(n^4)$ that can correct from $n-3$ insdel errors. Earlier constructions
required an exponential field size.
Joint work with Amir Shpilka and
Zachi Tamo.
Roni Con received the B.Sc. and M.Sc. degrees in applied
mathematics from Bar-Ilan University in 2012 and 2016, respectively. He is
currently pursuing a Ph.D. degree at the Computer Science Department, Tel
Aviv University, under the supervision of Amir Shpilka and Itzhak Tamo. H
is research interests include error-correcting codes and their application
s to theoretical computer science, DNA-based storage, and distributed stor
age systems.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 601
UID:123se24012024103060
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20221212T143000
DTEND;TZID="Asia/Jerusalem":20221212T153000
DTSTAMP;TZID="Asia/Jerusalem":20221212T143000
FREEBUSY;FBTYPE=BUSY:20221212T143000/20221212T153000
SUMMARY;LANGUAGE=en-US:colloq talk by George Varghese (UCLA) about CS Coll
oquia: Formal Methods for a Robust Network Ecosystem at 2022-12-12 14:30:0
0
DESCRIPTION;LANGUAGE=en-US:Network verification, applying formal methods t
o verify properties of router configurations, is already mainstream with s
tartups like Forward and Veriflow Networks, and divisions in established c
ompanies such as Amazon’s ARG and Cisco’s Candid. In this talk, I will su
rvey what remains to be done including: extending formal methods to implem
entations, and to other parts of the network ecosystem besides routing. I
will illustrate these points with recent work we have done on improving t
he robustness of the Domain Name Service (DNS). Notable errors in DNS hav
e rendered popular services such as GitHub, Twitter, and HBO inaccessible
for extended periods.
First, I will present GRoot (SIGCOMM 2020, Best
Student Paper), a new verification tool that performs exhaustive and proa
ctive static analysis of DNS configuration files (zone files) based on an
efficient algorithm to determine the equivalence class of all possible que
ries to guarantee key correctness properties. Next, I will describe a new
technique, SCALE (NSDI 2022), for finding RFC compliance errors in DNS nam
eserver implementations via symbolic execution of a DNS formal model to jo
intly generate test queries and zone files. Using SCALE, we identified 30
new bugs in 8 popular open-source DNS implementations such as BIND, PowerD
NS, KNOT, and NSD, including 3 previously unknown critical security vulner
abilities.
This talk is based on Siva Kakarla’s Ph.D. work at UCLA an
d is joint work with Ryan Beckett, Behnaz Armani at MSR , and Todd Millste
in at UCLA.
Bio:
George Varghese is the Jonathan B. Postel Professo
r of Networking in the Computer Science department at UCLA. He was electe
d to the National Academy of Engineering in 2017, to the National Academy
of Inventors in 2020, and to the Internet Hall of Fame in 2021, and to the
American Academy of Arts and Sciences in 2022.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Room 1003, EE Meyer Building
UID:123se24012024103200
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20221213T110000
DTEND;TZID="Asia/Jerusalem":20221213T120000
DTSTAMP;TZID="Asia/Jerusalem":20221213T110000
FREEBUSY;FBTYPE=BUSY:20221213T110000/20221213T120000
SUMMARY;LANGUAGE=en-US:msc talk by Daniel Shats about Patient-level Micros
atellite Status Assessment from Whole Slide Images By Combining Momentum C
ontrast Learning and Group Patch Embeddings at 2022-12-13 11:00:00
DESCRIPTION;LANGUAGE=en-US:Assessing microsatellite stability status of a
patient's colorectal cancer is crucial in personalizing treatment regime.
Recently, convolutional-neural-networks (CNN) combined with transfer-learn
ing approaches were proposed to circumvent traditional laboratory testing
for determining microsatellite status from hematoxylin and eosin stained b
iopsy whole slide images (WSI). However, the high resolution of WSI practi
cally prevent direct classification of the entire WSI. Current approaches
bypass the WSI high resolution by first classifying small patches extracte
d from the WSI, and then aggregating patch-level classification logits to
deduce the patient-level status. Such approaches limit the capacity to cap
ture important information which resides at the high resolution WSI data.
We introduce an effective approach to leverage WSI high resolution in
formation by momentum contrastive learning of patch embeddings along with
training a patient-level classifier on groups of those embeddings. Our app
roach achieves up to 7.4\% better accuracy compared to the straightforwa
rd patch-level classification and patient level aggregation approach with
a higher stability (AUC, $0.91 \pm 0.01$ vs. $0.85 \pm 0.04$, p-value$<0.0
1$). Our code can be found at \url{https://github.com/TechnionComputationa
lMRILab/colorectal_cancer_ai}.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture: 95409713968 and Taub 601
UID:123se24012024103270
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20221213T143000
DTEND;TZID="Asia/Jerusalem":20221213T153000
DTSTAMP;TZID="Asia/Jerusalem":20221213T143000
FREEBUSY;FBTYPE=BUSY:20221213T143000/20221213T153000
SUMMARY;LANGUAGE=en-US:colloq talk by Amir Globerson (Tel-Aviv university)
about CS Colloquia: Notions of Simplicity In Deep Learning: From Time Ser
ies to Images at 2022-12-13 14:30:00
DESCRIPTION;LANGUAGE=en-US:It is standard practice in deep learning to tra
in large models on relatively small datasets. This can potentially lead to
severe overfitting, but more often than not, test error is actually good.
This phenomenon has prompted research on the so-called "Implicit Bias of
Deep Learning Algorithms". Here I will discuss our recent works on multipl
e novel facets of this bias, and present theoretical and empirical results
in different settings. In particular, I will discuss analysis of implicit
bias in fine-tuning of large models, in learning temporal models (e.g., R
NNs) and in labeling images with very few examples.
Bio:
Prof. Glob
erson received his BSc in computer science and physics in 1997 from the He
brew University, and his PhD in computational neuroscience from the Hebrew
University in 2006. After his PhD, he was a postdoctoral fellow at the Un
iversity of Toronto and a Rothschild postdoctoral fellow at MIT. He joined
the Hebrew University school of computer science in 2008, and moved to th
e Tel Aviv University School of Computer Science in 2015. He was an associ
ate editor for the Journal of Machine Learning Research, and the Associate
Editor in Chief for the IEEE Transactions on Pattern Analysis and Machine
Intelligence. His work has received several paper awards (at NIPS, UAI, a
nd ICML). He also serves as Research Scientist at Google in Tel Aviv. In 2
018 he served as program co-chair for the UAI conference, and in 2019 he w
as the general co-chair for UAI in Tel Aviv. In 2019 he received the ERC c
onsolidator grant.
Host: Nir Rosenfeld.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Room 337 Taub Bld.
UID:123se24012024103210
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20221214T110000
DTEND;TZID="Asia/Jerusalem":20221214T120000
DTSTAMP;TZID="Asia/Jerusalem":20221214T110000
FREEBUSY;FBTYPE=BUSY:20221214T110000/20221214T120000
SUMMARY;LANGUAGE=en-US:msc talk by Idan Eldar about Direct Access to Answe
rs of Conjunctive Queries with Aggregation at 2022-12-14 11:00:00
DESCRIPTION;LANGUAGE=en-US:We study the fine-grained complexity of conjunc
tive queries with grouping and aggregation. For some common aggregate func
tions (e.g., min, max, sum), such a query can be phrased as an ordinary co
njunctive query over a database annotated with a suitable commutative semi
ring. Specifically, we study the ability to evaluate such queries by const
ructing in log-linear time a data structure that provides logarithmic-time
direct access to the answers ordered by a given lexicographic order. Thi
s task is nontrivial since the number of answers might be larger than log-
linear in the size of the input, and so, the data structure needs to provi
de a compact representation of the space of answers.
In the absence
of aggregation and annotation, past results provide a full classification
of the feasible and infeasible cases (queries and orders). We show that al
l past results continue to hold for annotated databases, assuming that the
annotation itself is not part of the lexicographic order. On the other ha
nd, we show infeasibility (under conventional complexity assumptions) for
the case of count-distinct that does not have any efficient representation
as a commutative semiring. We then turn to study the ability to include t
he aggregate function (or the annotation) in the lexicographic order. Amon
g the hardness results, standing out as tractable is the case of a semirin
g with an idempotent addition, such as those of min and max. Notably, this
case captures also count-distinct over a logarithmic-size domain.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture: 99541485114 and Taub 301
UID:123se24012024103260
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20221214T123000
DTEND;TZID="Asia/Jerusalem":20221214T133000
DTSTAMP;TZID="Asia/Jerusalem":20221214T123000
FREEBUSY;FBTYPE=BUSY:20221214T123000/20221214T133000
SUMMARY;LANGUAGE=en-US:Theory Seminar talk by Ronen Shaltiel (Haifa univer
sity) about Theory Seminar: Error Correcting codes that Achieve BSC Capaci
ty Against Channels that are Poly-Size Circuits at 2022-12-14 12:30:00
DESCRIPTION;LANGUAGE=en-US:Guruswami and Smith (J. ACM 2016) considered co
des for channels that are computationally bounded and flip at most a p-fra
ction of the bits of the codeword. This class of channels is significantly
stronger than Shannon’s binary symmetric channel (which flips each bit in
dependently with probability p) but weaker than Hamming’s channels (which
may flip at most a p-fraction of the bits, and are computationally unbound
ed).
The goal of this area is to construct codes against channels tha
t are computationally bounded (e.g., bounded memory channels, or channels
that are poly-size circuits). In this talk I will explain this direction,
focusing on a recent result by Shaltiel and Silbak (FOCS 2022) that consid
er channels that can be implemented by poly-size circuits.
The main
result of this work is a code that:
- Achieves optimal rate of 1-H(p) (m
atching the capacity of binary symmetric channels, and beating the capacit
y of Hamming channels).
- Has poly-time encoding and decoding algorithms
, after a randomized joint pre-processing stage (this is often referred to
as a "Monte-Carlo construction").
Our techniques rely on ideas from
coding theory, pseudorandomness and cryptography.
This is a joint wor
k with Jad Silbak.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 201
UID:123se24012024103280
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20221214T133000
DTEND;TZID="Asia/Jerusalem":20221214T143000
DTSTAMP;TZID="Asia/Jerusalem":20221214T133000
FREEBUSY;FBTYPE=BUSY:20221214T133000/20221214T143000
SUMMARY;LANGUAGE=en-US:pixel-club talk by Wolfgang Heidrich (KAUST Visual
Computing Center) about Pixel Club: Improving Computational Imaging System
s through Deep Learning and Optimization at 2022-12-14 13:30:00
DESCRIPTION;LANGUAGE=en-US:Computational imaging systems are based on the
joint design of optics and associated image reconstruction algorithms. His
torically, many such systems have employed simple transform-based reconstr
uction methods. Modern optimization methods and priors can drastically imp
rove the reconstruction quality in computational imaging systems. Furtherm
ore, learning-based methods can be used to design the optics along with th
e reconstruction method, yielding truly end-to-end optimized imaging syste
ms that outperform classical solutions.
Bio:
Wolfgang Heidrich is a
Professor of Computer Science and Electrical and Computer Engineering. Fr
om 2014 to 2021 he served as the director of the Visual Computing Center a
t KAUST, after 13 years as a faculty member at the University of British C
olumbia. He received his Ph.D. in from the University of Erlangen in 1999
and then worked as a Research Associate in the Computer Graphics Group of
the Max-Planck-Institute for Computer Science in Saarbrucken, Germany, bef
ore joining UBC in 2000. Prof. Heidrich's research interests lie at the in
tersection of imaging, optics, computer vision, computer graphics, and inv
erse problems. His more recent interest is in computational imaging, focus
ing on hardware-software co-design of the next generation of imaging syste
ms, with applications such as High-Dynamic Range imaging, compact computat
ional cameras, and hyperspectral cameras, to name just a few. Prof. Heidr
ich's work on High Dynamic Range Displays served as the basis for the tech
nology behind Brightside Technologies, which was acquired by Dolby in 2007
. Prof. Heidrich is a Fellow of the IEEE and Eurographics, and the recipie
nt of a Humboldt Research Award.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Room 815, EE Meyer Building
UID:123se24012024103290
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20221215T113000
DTEND;TZID="Asia/Jerusalem":20221215T123000
DTSTAMP;TZID="Asia/Jerusalem":20221215T113000
FREEBUSY;FBTYPE=BUSY:20221215T113000/20221215T123000
SUMMARY;LANGUAGE=en-US:msc talk by Guy Ohayon about Reasons for the Superi
ority of Stochastic Estimators over Deterministic Ones: Robustness, Consis
tency and Perceptual Quality at 2022-12-15 11:30:00
DESCRIPTION;LANGUAGE=en-US:Stochastic restoration algorithms allow to expl
ore the space of solutions that correspond to the degraded input. In this
paper we reveal additional fundamental advantages of stochastic methods ov
er deterministic ones, which further motivate their use. First, we prove t
hat any restoration algorithm that attains perfect perceptual quality and
whose outputs are consistent with the input must be a posterior sampler, a
nd is thus required to be stochastic. Second, we illustrate that while det
erministic restoration algorithms may attain high perceptual quality, this
can be achieved only by filling up the space of all possible source image
s using an extremely sensitive mapping, which makes them highly vulnerable
to adversarial attacks. Indeed, we show that enforcing deterministic mode
ls to be robust to such attacks profoundly hinders their perceptual qualit
y, while robustifying stochastic models hardly influences their perceptual
quality, and improves their output variability. These findings provide a
motivation to foster progress in stochastic restoration methods, paving th
e way to better recovery algorithms.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture: 2049693728
UID:123se24012024103160
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20221218T143000
DTEND;TZID="Asia/Jerusalem":20221218T153000
DTSTAMP;TZID="Asia/Jerusalem":20221218T143000
FREEBUSY;FBTYPE=BUSY:20221218T143000/20221218T153000
SUMMARY;LANGUAGE=en-US:Coding_Theory_Semina talk by Nathan Lindzey (CS, Te
chnion) about Coding Theory: Association Schemes and Injection Codes at 20
22-12-18 14:30:00
DESCRIPTION;LANGUAGE=en-US:The first half of the talk will overview Delsar
te's association scheme-theoretic approach to coding theory, which has bee
n used extensively over the past few decades to obtain bounds on many diff
erent classes of codes. For example, the Hamming scheme and the Johnson sc
heme both have classical roles in the theory of linear codes and constant
-weight codes respectively, whereas the permutation scheme has been used r
ecently in the study of permutation codes, a less well-known class of code
s defined over permutations that have applications to powerline communicat
ion design. The last half of the talk will focus on a simultaneous general
ization of the Johnson scheme and the permutation scheme called the inject
ion scheme, an association scheme defined over the collection of all injec
tive functions from a k-element set to a n-element set where k is no large
r than n. Through a deeper understanding of the algebraic structure of thi
s scheme, we are able to improve the known bounds for injection codes for
various values of k and n. This is joint work with Peter Dukes and Ferdina
nd Ihringer.
Nathan Lindzey received his Ph.D from the Department o
f Combinatorics and Optimization at the University of Waterloo under the s
upervision of Chris Godsil and Joseph Cheriyan. He is currently a postdoc
under the supervision of Yuval Filmus in the Faculty of Computer Science a
t Technion.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 601
UID:123se24012024103330
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20221220T103000
DTEND;TZID="Asia/Jerusalem":20221220T113000
DTSTAMP;TZID="Asia/Jerusalem":20221220T103000
FREEBUSY;FBTYPE=BUSY:20221220T103000/20221220T113000
SUMMARY;LANGUAGE=en-US:colloq talk by Yonathan Efroni (Meta, New York) abo
ut CS Lecture: Reinforcement Learning in the Presence of Irrelevant Inform
ation at 2022-12-20 10:30:00
DESCRIPTION;LANGUAGE=en-US:Reinforcement Learning (RL) is a field concerne
d with designing general purpose learning algorithms that solve sequential
-decision tasks. In recent years, by using deep neural networks, RL algori
thms were applied on high-dimensional and challenging domains, witnessing
unprecedented success. Yet, despite recent advancements, the theoretical f
oundations of high-dimensional RL are not fully understood.
A recurr
ing theme in high-dimensional RL is the presence of irrelevant information
in the observations. E.g., in a visual navigation task the observation m
ight capture the movement of clouds, which is irrelevant for reaching the
goal location. This calls for natural questions: Can such tasks be learned
efficiently, depending only on the complexity of the relevant information
? Can RL algorithms be robust to noise in observations? Surprisingly, cont
emporary RL algorithms may provably fail in the presence of irrelevant inf
ormation.
In this talk, I will elaborate on these failure cases and p
resent our new provable approaches for high-dimensional RL with irrelevant
information. Shared to these are techniques to filter the irrelevant info
rmation while guaranteeing near-optimal behavior. I will conclude with ex
perimental results showcasing challenges and solutions in practice.
Bio:
Yonathan is a research scientist at Meta. Prior to that he complete
d his post-doctorate in Microsoft Research, New York. He obtained his PhD
from the Technion, advised by Prof. Shie Mannor, and his Master from the W
eizmann institute in Physics. His work won the outstanding paper award in
AAAI19 and a best paper award in the OptRL workshop in NeurIPS19.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 601
UID:123se24012024103320
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20221227T113000
DTEND;TZID="Asia/Jerusalem":20221227T123000
DTSTAMP;TZID="Asia/Jerusalem":20221227T113000
FREEBUSY;FBTYPE=BUSY:20221227T113000/20221227T123000
SUMMARY;LANGUAGE=en-US:colloq talk by Shafi Goldwasser (UC Berkeley) about
CS Colloquia: Constructing and deConstructing Trust in ML: A new Role for
Cryptography Today at 2022-12-27 11:30:00
DESCRIPTION;LANGUAGE=en-US:For decades now cryptographic tools and models
have at their essence transformed technology platforms controlled by worst
case adversaries to trustworthy platforms. In this talk I will describe h
ow to use cryptographic tools and cryptographic modeling to build trust in
various phases of the machine learning pipelines. We will touch on privac
y in the training and inference stage, verification protocols for the qual
ity of machine learning models, and robustness in presence of adversaries.
If time permits, we will show how cryptographic tools can be brought to b
uild trust in the legal domain.
Short Bio:
Shafi Goldwasser is Di
rector of the Simons Institute for the Theory of Computing, and Professor
of Electrical Engineering and Computer Science at the University of Califo
rnia Berkeley. Goldwasser is also Professor of Electrical Engineering and
Computer Science at MIT and Professor of Computer Science and Applied Math
ematics at the Weizmann Institute of Science, Israel. Goldwasser holds a B
.S. Applied Mathematics from Carnegie Mellon University (1979), and M.S. (
1981) and Ph.D. in Computer Science from the University of California Berk
eley (1984).
Goldwasser's pioneering contributions include the introd
uction of probabilistic encryption, interactive zero knowledge protocols,
elliptic curve primality testing, hardness of approximation proofs for com
binatorial problems, and combinatorial property testing.
Goldwasser
was the recipient of the ACM Turing Award in 2012, the Gödel Prize in 1993
and in 2001, the ACM Grace Murray Hopper Award in 1996, the RSA Award in
Mathematics in 1998, the ACM Athena Award for Women in Computer Science in
2008, the Benjamin Franklin Medal in 2010, the IEEE Emanuel R. Piore Awar
d in 2011, the Simons Foundation Investigator Award in 2012, the BBVA Foun
dation Frontiers of Knowledge Award in 2018, and the L’Oréal-UNESCO For Wo
men in Science Award in 2021. Goldwasser is a member of the NAS, NAE, AAAS
, the Russian Academy of Science, the Israeli Academy of Science, and the
London Royal Mathematical Society. Goldwasser holds honorary degrees from
Ben Gurion University, Bar Ilan University, Carnegie Mellon University, Ha
ifa University, University of Oxford, and the University of Waterloo, and
has received the UC Berkeley Distinguished Alumnus Award and the Barnard C
ollege Medal of Distinction.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Room 337 Taub Bld.
UID:123se24012024103310
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20221228T113000
DTEND;TZID="Asia/Jerusalem":20221228T123000
DTSTAMP;TZID="Asia/Jerusalem":20221228T113000
FREEBUSY;FBTYPE=BUSY:20221228T113000/20221228T123000
SUMMARY;LANGUAGE=en-US:ceClub talk by Ben Nassi (Ben-Gurion University) ab
out ceClub: Finding darkness in the Light: Recovering Speech and Cryptogra
phic keys from Light Emitted from Power LEDs and Light Bulbs at 2022-12-28
11:30:00
DESCRIPTION;LANGUAGE=en-US:In this talk, I will present a journey that sta
rted three years ago at the intersection between light leakage and informa
tion confidentiality. In the first part of the talk, I will present the to
pic of electro-optical speech eavesdropping which is based on three method
s we developed to recover speech from light emitted from light bulbs (Lamp
hone USENIX Security 22), power LEDs of speakers (Glowworm Attack CCS 21),
and from light reflected from shiny ornaments and objects (The Little Sea
l Bug - BH Asia 22). We will discuss the threat model and its significance
with respect to related works, and hear speech recoveries from light meas
urements (obtained by a photodiode) 25-35 meters away.
In the second
part of the talk, I will present visual cryptanalysis, a new method to re
cover secret keys of three different cryptosystems (RSA, ECDSA, and SIKE)
by obtaining optical traces from power LEDs of various devices (Galaxy S8,
card reader, Raspberry Pi) that run common cryptographic libraries (GnuPG
, Libgcryipt, and PQCrypto-SIDH), using a photodiode located 25 meters awa
y. We will discuss the new threat model and its significance with respect
to related works, and understand the origin of the vulnerability.
At
the end of the talk, I will present an ECDSA key recovery of a smart card
using a video taken by a security camera placed 2.5 meters from the power
LED of a card reader by exploiting the rolling shutter of video cameras.
Bio:
Dr. Ben Nassi is a postdoctoral researcher in the Software an
d Information Systems Engineering department at the Ben-Gurion University
of the Negev and soon to be a postdoctoral researcher at Cornell Tech. He
is interested in building robust systems and investigating the security an
d privacy of cyber-physical systems (drones, semi-autonomous cars) and har
dware/devices (microcontrollers, smartphones, smart cards) in the topics o
f side-channel attacks, AI robustness, and applied cryptography using sign
al processing and machine learning techniques. His research has been prese
nted at top academic conferences (S&P, CCS, USENIX Security, UbiComp) publ
ished in journals (IEEE TIFS, MDPI Sensors) and Magazines (IEEE Computer,
Communications of the ACM), and covered by international media (Wired, Ars
Technica, Motherboard, Forbes, The Washington Post, Bloomberg, Business In
sider). Ben has spoken at prestigious industrial conferences (Black Hat As
ia and USA, RSAC USA, AI Week, CodeBlue, SecTor, and CyberTech) and he ser
ves as a PC member in ACM CCS (22 and 23) and BlackHat (22 and 23). His re
search entitled him to the BGU Dean Award for excellence in Ph.D., and two
nominations for the Pwnie Award.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Room 861, EE Meyer Building & Zoom Lecture: 94673013539
UID:123se24012024103380
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20221228T123000
DTEND;TZID="Asia/Jerusalem":20221228T143000
DTSTAMP;TZID="Asia/Jerusalem":20221228T123000
FREEBUSY;FBTYPE=BUSY:20221228T123000/20221228T143000
SUMMARY;LANGUAGE=en-US:CSpecial Event about CS RESEARCH DAY 2022 at 2022-
12-28 12:30:00
DESCRIPTION;LANGUAGE=en-US:The 10th CS Research Day for graduate studies w
ill be held on Wednesday, December 28, 2022 between 12:30-14:30, at the lo
bby of the CS Taub Building.
Research Day events are opportunity for
our graduate students to expose their researches using posters and present
ations to CS faculty and all degrees students, Technion distinguished repr
esentatives and to high-ranking delegates from the hi-tech leading industr
y companies in Israel and abroad.
The participating researches will
be on various topics: Cryptology and Cyber, Data Centers and Clouds, Graph
ics, Intelligent Systems and Scientific Computation, Machine Learning and
Information Retrieval, Systems and Applications, Testing and Verification,
Theory of Computer Science.
Students wishing to present their res
earch (posters should size 98x68 cm, preferably vertically, and be submitt
ed in Pdf format), are kindly requested to register here.
The presenting researches
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:CS Taub Lobby
UID:123se24012024103030
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20230102T173000
DTEND;TZID="Asia/Jerusalem":20230102T183000
DTSTAMP;TZID="Asia/Jerusalem":20230102T173000
FREEBUSY;FBTYPE=BUSY:20230102T173000/20230102T183000
SUMMARY;LANGUAGE=en-US:CSpecial Event about Recruitment Day by Microsoft
at 2023-01-02 17:30:00
DESCRIPTION;LANGUAGE=en-US:You are invited to Microsoft Recruitment day, t
o a question and answer session with the team about the company's recr
uitment process, and to a lecture by Noa Berman, a software developer at M
icrosoft Security, on: Ransomware attacks and how we defend against them a
t Microsoft Defender for Endpoint, on Monday, January 2, 17:30, at Taub 9.
Please pre-register.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 9
UID:123se24012024103400
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20230103T103000
DTEND;TZID="Asia/Jerusalem":20230103T123000
DTSTAMP;TZID="Asia/Jerusalem":20230103T103000
FREEBUSY;FBTYPE=BUSY:20230103T103000/20230103T123000
SUMMARY;LANGUAGE=en-US:CSpecial Event about Yahoo Research at CS at 2023-
01-03 10:30:00
DESCRIPTION;LANGUAGE=en-US:Yahoo Research will visit CS
for a special meeting with graduate stud
ents on Tuesday, January 3, 2023 starting at 10:
30 at the Grads Club, 2nd floor (at the end of t
he corridor), Taub Computer Science Building:
Program:
10:30 - Ga
thering
11:00 - Intro - Yahoo Israel Research Center
11:15 - Lecture
1: Leveraging User Email Actions to Improve Ad-Close Prediction - by Yaro
slav Fyodorov
11:35 - Lecture 2: Augmentation for Consistent Categoriza
tion - by Stav Yanovsky Daye
11:55 - Lecture 3: Consistent Text Categor
ization with Data Augmentation - by Alex Shtoff
12:00 - Mingling
Please register in adv
ance
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Grads Club, CS Taub 2nd floor
UID:123se24012024103480
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20230103T110000
DTEND;TZID="Asia/Jerusalem":20230103T120000
DTSTAMP;TZID="Asia/Jerusalem":20230103T110000
FREEBUSY;FBTYPE=BUSY:20230103T110000/20230103T120000
SUMMARY;LANGUAGE=en-US:colloq talk by Brit Youngmann (CSAIL MIT) about CS
Lecture: Data Tools for Accelerated Scientific Discoveries at 2023-01-03 1
1:00:00
DESCRIPTION;LANGUAGE=en-US:Causal inference is fundamental to empirical re
search in natural and social sciences and is essential for scientific disc
overies. Two key challenges for conducting causal inference are (i) acquir
ing all attributes required for the analysis, and (ii) identifying which a
ttributes should be included in the analysis. Failing to include all neces
sary attributes may lead to false discoveries and erroneous conclusions. H
owever, in real-world settings, analysts may only have access to partial d
ata. Further, to identify which attributes should be included in the analy
sis, 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 work
s that address these challenges by leveraging data management techniques a
nd ideas.
Bio:
Brit is a postdoc researcher at CSAIL MIT, working
with Prof. Michael Cafarella. She received her Ph.D. at Tel-Aviv Universit
y under the supervision of Prof. Tova Milo. Her research is centered aroun
d informative and responsible data science and causal analysis. Brit is th
e recipient of several awards, including the data science fellowship for o
utstanding Ph.D. students of the planning and budgeting committee of the I
sraeli council for higher education (VATAT), the Schmidt postdoctoral awar
d for women in mathematical and computing sciences, and the planning and b
udgeting committee of the Israeli council for higher education (VATAT) pos
tdoctoral scholarship in Data Science. She served on multiple program comm
ittees, including at the SIGMOD and ICDE conferences.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 601
UID:123se24012024103470
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20230103T143000
DTEND;TZID="Asia/Jerusalem":20230103T153000
DTSTAMP;TZID="Asia/Jerusalem":20230103T143000
FREEBUSY;FBTYPE=BUSY:20230103T143000/20230103T153000
SUMMARY;LANGUAGE=en-US:colloq talk by Michael Lustig (UC Berkeley) about C
S Colloquia: Adventures in Computational MRI at 2023-01-03 14:30:00
DESCRIPTION;LANGUAGE=en-US:Magnetic resonance imaging (MRI) is a powerful,
ionizing-radiation-free medical imaging modality. The vast physical and p
hysiological parameters, which MRI is sensitive to, makes it possible to v
isualize both structure and function in the body. However the prolonged ti
me necessary to capture the information in this large parameter space rema
ins a major limitation of this phenomenal modality, which the field of com
putational 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 d
escribe some of the efforts my group has been engaged in towards mitigatin
g with motion and dynamics that occurs during MRI scanning, in particular
when performing body imaging of pediatric patients. Specifically I will fo
cus on unsupervised and supervised methods for dynamic 2D and 3D imaging a
nd learning based high fidelity reconstructions of fine structures and te
xtures.
Short bio:
Michael (Miki) Lustig is a Professor in Electri
cal Engineering and Computer Science. He joined the faculty of the EECS D
epartment at UC Berkeley in Spring 2010. He received his B.Sc. in Electric
al Engineering from the Technion, Israel Institute of Technology in 2002.
He received his MSc and Ph.D. in Electrical Engineering from Stanford Univ
ersity in 2004 and 2008, respectively. His research focuses on computation
al imaging methods in medical imaging, particularly Magnetic Resonance Ima
ging (MRI)— these include a spectrum of work ranging from Hardware, throug
h 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:123se24012024103360
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20230104T103000
DTEND;TZID="Asia/Jerusalem":20230104T113000
DTSTAMP;TZID="Asia/Jerusalem":20230104T103000
FREEBUSY;FBTYPE=BUSY:20230104T103000/20230104T113000
SUMMARY;LANGUAGE=en-US:colloq talk by Yoav Levine (AI21 Labs) about CS Lec
ture: Theoretical and practical principles for designing, training, and de
ploying huge language models at 2023-01-04 10:30:00
DESCRIPTION;LANGUAGE=en-US: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 arch
itecture 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 t
hrusts in contemporary NLP, aimed to serve as timely constructive feedback
for the strong empirical pull in this field.
I will begin by descri
bing our theoretical framework for analyzing Transformers, and present res
ults on the depth to width tradeoffs in Transformers, identified bottlenec
ks within internal Transformer dimensions, and identified biases introduce
d during the Transformer self-supervised pretraining phase. This framework
has guided the design and scale of several of the largest existing langua
ge 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 or
der to increase efficiency of the self-supervised pretraining phase. Subse
quently, I will describe novel principles for addressing a present-day pro
blem stemming from the above success of scaling, namely, how to deploy a h
uge language model such that it specializes in many different use cases si
multaneously (e.g., supporting many different customer needs simultaneousl
y). Finally, I will comment on future challenges in this field, and will r
elatedly present a recent theoretical result on the importance of intermed
iate supervision when solving composite NLP tasks.
This talk is base
d on works published in NeurIPS 2020, ACL 2020, ICLR 2021 (spotlight), ICM
L 2021, ICLR 2022 (spotlight), ICML 2022 (workshop), as well as on several
recent preprints.
Bio:
Dr. Yoav Levine serves as co-Chief Scientis
t 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. H
is PhD studies were supported by the Israeli Academy of Sciences Adams fel
lowship, and for them he has received the Blavatnik PhD Prize awarded to t
he 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 phys
ics 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 e
xcellence program.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 601
UID:123se24012024103460
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20230104T123000
DTEND;TZID="Asia/Jerusalem":20230104T133000
DTSTAMP;TZID="Asia/Jerusalem":20230104T123000
FREEBUSY;FBTYPE=BUSY:20230104T123000/20230104T133000
SUMMARY;LANGUAGE=en-US:CSpecial Event about Recruitment Day and Workshop
by Intel at 2023-01-04 12:30:00
DESCRIPTION;LANGUAGE=en-US:Intel will hold a recruitmen
t day and will present employment opportu
nities, as well as a "Fusion 360" workshop of 3D modeling, print
ing and Makers experie
nce, on Wednesday, January 4, 2023, starting at
12:30 in the Taub lobby.
For the workshop please pre-register in advance (the number of pl
aces for the workshop is limited and subject to confirmation of registrati
on)
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:CS Taub Lobby
UID:123se24012024103420
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20230104T123000
DTEND;TZID="Asia/Jerusalem":20230104T133000
DTSTAMP;TZID="Asia/Jerusalem":20230104T123000
FREEBUSY;FBTYPE=BUSY:20230104T123000/20230104T133000
SUMMARY;LANGUAGE=en-US:Theory Seminar talk by Ilya Volkovich (Boston Colle
ge) about Theory Seminar: Mutual Empowerment between Circuit Obfuscation a
nd Circuit Minimization at 2023-01-04 12:30:00
DESCRIPTION;LANGUAGE=en-US:We study close connections between Indistinguis
hability 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:
If there exists a perfect (imperfect) I
O that is computationally-secure against non-uniform polynomial-size circu
its, then we obtain fixed-polynomial lower bounds against NP(MA).
In
addition, computationally-secure IO against non-uniform polynomial-size ci
rcuits imply super-polynomial lower bounds against NEXP.
If MCSP is i
n BPP, then statistical security and computational security for IO are equ
ivalent.
To the best of our knowledge, this is the first consequence
of strong circuit lower bounds from the existence of an IO. The results ar
e obtained via a construction of an optimal universal distinguisher, compu
table in randomized polynomial time with access to the MCSP oracle, that w
ill distinguish any two circuit-samplable distributions with the advantage
that is the statistical distance between these two distributions minus so
me negligible error term. This is our main technical contribution. As anot
her 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:123se24012024103440
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20230105T143000
DTEND;TZID="Asia/Jerusalem":20230105T153000
DTSTAMP;TZID="Asia/Jerusalem":20230105T143000
FREEBUSY;FBTYPE=BUSY:20230105T143000/20230105T153000
SUMMARY;LANGUAGE=en-US:msc talk by Amit Ganz about Online Submodular Welfa
re Maximization with General Utilities at 2023-01-05 14:30:00
DESCRIPTION;LANGUAGE=en-US:We consider the online Submodular Welfare probl
em.
In this problem we are given n bidders each equipped with a submo
dular utility and m items that arrive online.
The goal is to assign
each item, once it arrives, to a bidder or discard it, while maximizing th
e sum of utilities.
The case of monotone utilities has attracted much
attention, however much less is known once utilities are general and not
necessarily monotone.
When an adversary determines the items' arrival
order, we present an algorithm, inspired by the algorithm of [Dobzinski-S
chapira SODA`06], that achieves a competitive ratio of (n/(8n-4)).
Fo
r a single bidder, this ratio equals 1/4 and it gracefully degrades to 1/8
as the number of bidders increases.
We note that for a single bidder
, online Submodular Welfare is equivalent to online Unconstrained Submodul
ar Maximization, for which a hardness of 1/4 is known alongside an algorit
hm with a matching competitive ratio.
To the best of our knowledge, n
o 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:123se24012024103410
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20230110T103000
DTEND;TZID="Asia/Jerusalem":20230110T113000
DTSTAMP;TZID="Asia/Jerusalem":20230110T103000
FREEBUSY;FBTYPE=BUSY:20230110T103000/20230110T113000
SUMMARY;LANGUAGE=en-US:colloq talk by Omri Ben-Eliezer (MIT) about CS Lect
ure: Fast Algorithms for Complex Environments at 2023-01-10 10:30:00
DESCRIPTION;LANGUAGE=en-US:Our modern life is marked by continuous interac
tion with huge and complex computational environments, a setting which giv
es rise to numerous theoretical and algorithmic challenges. Algorithms now
adays are often required to optimize objectives that may be theoretically
ill-defined, on big data that is complex-structured, while maintaining com
putational efficiency and provable guarantees such as privacy and robustne
ss. In this talk I will discuss some of my work developing new computation
al models and fast (e.g., sublinear-time or sublinear-space) algorithms fo
r these modern settings, where data is highly structured or undergoes comp
lex dynamics. I will focus on three representative lines of work: (i) the
first systematic investigation of adversarial robustness in streaming algo
rithms, (ii) a new algorithmic framework for real-world social networks ba
sed on core-periphery sparsification, and (iii) active learning and testin
g the “geometry of data” in low-dimensional settings. Through these exampl
es, I will demonstrate how the symbiosis between modeling and algorithm de
sign often leads naturally to new structural insights and multidisciplinar
y connections.
Bio:
Omri Ben-Eliezer is an instructor (postdoc)
in applied mathematics at MIT. He received his PhD in computer science fr
om 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 specifi
c interests including sublinear-time and streaming algorithms, large netwo
rks, robustness and privacy, and knowledge representation. For his work, O
mri received several awards, including best paper awards at PODS 2020 and
at CVPR 2020 Workshop on Text and Documents, the 2021 SIGMOD Research High
light Award, and the first Blavatnik Prize for outstanding Israeli doctora
l students in computer science.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 601
UID:123se24012024103510
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20230110T113000
DTEND;TZID="Asia/Jerusalem":20230110T123000
DTSTAMP;TZID="Asia/Jerusalem":20230110T113000
FREEBUSY;FBTYPE=BUSY:20230110T113000/20230110T123000
SUMMARY;LANGUAGE=en-US:pixel-club talk by Yuval Bahat (Princeton & Univers
ity of Siegon) about Pixel Club: Neural Volume Super-Resolution at 2023-0
1-10 11:30:00
DESCRIPTION;LANGUAGE=en-US:Neural volumetric representations have become a
widely adopted model for radiance fields in 3D scenes. These representati
ons are fully implicit or hybrid function approximators of the instantaneo
us volumetric radiance in a scene, which are typically learned from multi-
view captures of the scene. We investigate the new task of neural volume s
uper-resolution - rendering high-resolution views corresponding to a scene
captured at low resolution. To this end, we propose a neural super-resolu
tion network that operates directly on the volumetric representation of th
e scene. This approach allows us to exploit an advantage of operating in t
he volumetric domain, namely the ability to guarantee consistent super-res
olution across different viewing directions. To realize our method, we dev
ise a novel 3D representation that hinges on multiple 2D feature planes. T
his 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 qu
antitatively and qualitatively on a diverse set of unseen 3D scenes, demon
strating a significant advantage over existing approaches.
Bio:
Yu
val holds a joint postdoctoral researcher position at the computational im
aging lab in Princeton and the ZESS center at the university of Siegen. Hi
s research interests lie at the intersection of computer vision and comput
ational photography with Machine learning. He was previously a postdoctora
l researcher at Prof. Tomer Michaeli's lab at the Technion, after completi
ng his PhD at the Weizmann Institute of Science, advised by Prof. Michal I
rani. Prior to that he completed his M.Sc. at the Technion with Prof. Yoav
Y. Schechner.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Room 337 taub bld.
UID:123se24012024103550
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20230110T123000
DTEND;TZID="Asia/Jerusalem":20230110T143000
DTSTAMP;TZID="Asia/Jerusalem":20230110T123000
FREEBUSY;FBTYPE=BUSY:20230110T123000/20230110T143000
SUMMARY;LANGUAGE=en-US:CSpecial Event about Recruitment Day by VAYYAR at
2023-01-10 12:30:00
DESCRIPTION;LANGUAGE=en-US:VAYYAR representatives will visit CS to present
products and developments and to offer open positions, on Wednesday, Janu
ary 11, 2023, 12:30, Taub lobby.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:CS Taub Lobby
UID:123se24012024103530
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20230110T143000
DTEND;TZID="Asia/Jerusalem":20230110T153000
DTSTAMP;TZID="Asia/Jerusalem":20230110T143000
FREEBUSY;FBTYPE=BUSY:20230110T143000/20230110T153000
SUMMARY;LANGUAGE=en-US:colloq talk by Tamir Tassa (Open University of Isra
el) about CS Colloquia: Fear Not, Vote Truthfully: Secure E-Voting Protoco
ls for Score-and Order-Based Rules at 2023-01-10 14:30:00
DESCRIPTION;LANGUAGE=en-US:Electronic voting systems are essential for hol
ding 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: namel
y, to certify that the computed results are consistent with the cast ballo
ts, and that the privacy of the voters is preserved. We propose secure vot
ing protocols for elections that are governed by two central families of v
oting rules: score-based and order-based rules. Our protocols offer perfec
t ballot secrecy, in the sense that they issue only the required output, w
hile no other information on the cast ballots is revealed. Such perfect se
crecy, 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 pri
vacy, and their lightweight nature, make them an adequate and powerful too
l for democracies of any size.
Joint work with Lihi Dery and Avishay
Yanai
Short bio:
Professor Tamir Tassa is a faculty member in the
Department of Mathematics and Computer Science at the Open University of I
srael. Previously, he served as a lecturer and researcher in the School of
Mathematical Sciences at Tel Aviv University, and in the Department of Co
mputer Science at Ben Gurion University. During the years 1993-1996 he ser
ved as an assistant professor of Computational and Applied Mathematics at
the University of California, Los Angeles. He earned his PhD in mathematic
s from Tel Aviv University in 1993. His recent research interests include
secure multiparty computation, privacy-preserving data publishing and data
mining, and secret sharing.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Room 337 taub bld.
UID:123se24012024103500
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20230110T183000
DTEND;TZID="Asia/Jerusalem":20230110T203000
DTSTAMP;TZID="Asia/Jerusalem":20230110T183000
FREEBUSY;FBTYPE=BUSY:20230110T183000/20230110T203000
SUMMARY;LANGUAGE=en-US:CSpecial Event about Open Source Workshop at 2023
-01-10 18:30:00
DESCRIPTION;LANGUAGE=en-US:You are invited to an Open Source workshop on o
pen source and how contributing to open source helps professional developm
ent and gaining experience at any stage of your career,
in a lecture by Michal Forg, front end developer at Gong company a
nd manager of the largest open source community in Israel, Pull Request, o
n Tuesday, January 10, 2023 at 18:30 in T
aub 337.
Please r
egister in advance
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Room 337 taub bld.
UID:123se24012024103520
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20230111T123000
DTEND;TZID="Asia/Jerusalem":20230111T133000
DTSTAMP;TZID="Asia/Jerusalem":20230111T123000
FREEBUSY;FBTYPE=BUSY:20230111T123000/20230111T133000
SUMMARY;LANGUAGE=en-US:Theory Seminar talk by Ilan Doron-Arad (CS, Technio
n) about Theory Seminar: Efficient approximation for budgeted matroid inde
pendent set, budgeted matching, and budgeted matroid intersection at 2023-
01-11 12:30:00
DESCRIPTION;LANGUAGE=en-US:Abstract: We consider the budgeted matroid inde
pendent set problem. The input is a ground set, where each element has a c
ost 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 t
otal profit subject to the matroid and budget constraints. Several well kn
own 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 slig
ht generalization to the multi-budgeted matroid independent set problem ha
s a PTAS but does not admit an efficient polynomial-time approximation sch
eme (EPTAS). This implies a PTAS for our problem, which is the best known
result prior to our work.
Our main contribution is an EPTAS for the b
udgeted matroid independent set problem, and a generalization of the techn
ique for obtaining an EPTAS for budgeted matching and budgeted matroid int
ersection. A key idea of the scheme is to find a representative set for th
e instance, whose cardinality depends solely on 1/\eps,
where \eps>0 is
the accuracy parameter of the scheme. Our scheme enumerates over subsets o
f the representative set and extends each subset using Lagrangian relaxati
on techniques.
For a single matroid, the representative set is identi
fied via a matroid basis minimization, which can be solved by a simple gre
edy approach. For matching and matroid intersection, matroid basis minimiz
ation is used as a baseline for a more sophisticated approach.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 201
UID:123se24012024103540
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20230111T163000
DTEND;TZID="Asia/Jerusalem":20230111T173000
DTSTAMP;TZID="Asia/Jerusalem":20230111T163000
FREEBUSY;FBTYPE=BUSY:20230111T163000/20230111T173000
SUMMARY;LANGUAGE=en-US:Coding_Theory_Semina talk by Prof. Camilla Hollanti
(Aalto University, Finland) about Coding Theory: Capacity of Private Info
rmation Retrieval from Coded and Colluding Servers at 2023-01-11 16:30:00
DESCRIPTION;LANGUAGE=en-US:Private information retrieval (PIR) addresses t
he question of how to retrieve data items from a database or cloud without
disclosing information about the identity of the data items retrieved. Th
e area has received renewed attention in the context of PIR from coded sto
rage. Here, the files are distributed over the servers according to a stor
age code instead of mere replication. Alongside with the basic principles
of PIR, we will review recent capacity results and demonstrate the usefuln
ess of the so-called star product PIR scheme.
The talk is based on jo
int work with Ragnar Freij-Hollanti, Oliver Gnilke, Lukas Holzbaur, David
Karpuk, and Jie Li.
Bio:
Camilla Hollanti received the M.Sc. and P
h.D. degrees from the University of Turku, Finland, in 2003 and 2009, resp
ectively, both in pure mathematics. Since 2011, she has been with the Depa
rtment of Mathematics and Systems Analysis, Aalto University, Finland, whe
re she currently works as a professor and the vice head of the department,
and leads a research group in Algebra, Number Theory, and Applications. F
rom 2017 to 2020, she was affiliated with the Institute of Advanced Studie
s, Technical University of Munich, where she held a Hans Fischer Fellowshi
p. Her research interests lie within applications of algebraic number theo
ry to wireless communications and physical layer security as well as in co
mbinatorial and coding theoretic methods related to secure and private com
putation.
Dr. Hollanti is a coauthor of over a hundred scientific pee
r-reviewed publications and is a recipient of several grants, including se
ven Academy of Finland Grants. In 2014, she received the World Cultural Co
uncil Special Recognition Award for young researchers. In 2017, the Finnis
h Academy of Science and Letters awarded her the Väisälä Prize in Mathemat
ics. Since 2020, she has been serving as a member of the Board of Governor
s of the IEEE Information Theory Society, and she was a general chair of t
he IEEE ISIT 2022. She is currently an Editor of the AIMS Journal on Advan
ces in Mathematics of Communications, SIAM Journal on Applied Algebra and
Geometry, IEEE Transactions on Information Theory, and Annales Fennici Mat
hematici.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture: 98686325633
UID:123se24012024103300
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20230112T103000
DTEND;TZID="Asia/Jerusalem":20230112T113000
DTSTAMP;TZID="Asia/Jerusalem":20230112T103000
FREEBUSY;FBTYPE=BUSY:20230112T103000/20230112T113000
SUMMARY;LANGUAGE=en-US:colloq talk by Re'em Harel (Head of Algorithms, NRC
N) about CS Guest Lecture: Automatic Parallelization for Concurrent Progra
mming – Past, Present, and Future at 2023-01-12 10:30:00
DESCRIPTION;LANGUAGE=en-US:Introducing parallelism to applications is a co
mplex and tedious task. As a result, the field named automatic paralleliza
tion emerged. Automatic parallelization refers to the seamless introductio
n 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 embedd
ed in compilers and dedicated source-to-source (S2S) compilers. However, r
ecent advances and success in deep Natural Language Processing (NLP) inspi
red models for similar code-related tasks. For example, Codex (based on GP
T) generates and suggests code. The possibility of creating a similar mode
l for automatically introducing, or at the very least suggesting, OpenMP d
irectives rises.
In this talk, we will go through this field's histor
y and the state-of-the-art - from the deterministic/algorithmic approach t
o the Machine Learning one, based on our recent publications.
Referen
ces:
● Harel, R. E., Mosseri, I., Levin, H., Alon, L. O., Rusanovsky,
M., & Oren, G. (2020). Source-to-source parallelization compilers for sci
entific shared-memory multi-core and accelerated multiprocessing: analysis
, pitfalls, enhancement and potential. International Journal of Parallel P
rogramming, 48(1), 1-31.
● Mosseri, I., Alon, L. O., Harel, R. E., & Ore
n, G. (2020, September). ComPar: optimized multi-compiler for automatic Op
enMP S2S parallelization. In International Workshop on OpenMP (pp. 247-262
). Springer, Cham.
● Harel, R. E., Pinter, Y., & Oren, G. (2022). Learni
ng 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 202
2)
Biography: 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 intro
duce parallelization schemes, such as OpenMP directives and MPI functions,
to new and legacy codes. In addition, he is a researcher in the scientifi
c computing lab at NRCN, focusing on parallel programming and scientific c
omputing.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture: 95409713968
UID:123se24012024103340
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20230112T123000
DTEND;TZID="Asia/Jerusalem":20230112T133000
DTSTAMP;TZID="Asia/Jerusalem":20230112T123000
FREEBUSY;FBTYPE=BUSY:20230112T123000/20230112T133000
SUMMARY;LANGUAGE=en-US:Theory Seminar talk by Merav Parter (Weizmann Insti
tute of Science) about Theory Seminar: A Graph Theoretic Approach for Resi
lient Distributed Algorithms at 2023-01-12 12:30:00
DESCRIPTION;LANGUAGE=en-US:Following the immense recent advances in distri
buted networks, the explosive growth of the Internet, and our increased de
pendency on these infrastructures, guaranteeing the uninterrupted operatio
n of communication networks has become a major objective in network algori
thms. The modern instantiations of distributed networks, such as the Bitco
in network and cloud computing, introduce new security challenges that des
erve urgent attention in both theory and practice.
In this talk, I wi
ll present a unified framework for obtaining fast, resilient and secure di
stributed algorithms for fundamental graph problems. Our approach is based
on a graph-theoretic perspective in which common notions of resilient req
uirements are translated into suitably tailored combinatorial graph struct
ures. We will discuss recent developments along the following two lines of
research:
– Initiating and establishing the theoretical exploration
of security in distributed graph algorithms. Such a notion has been addres
sed before mainly in the context of secure multi-party computation (MPC).
The heart of our approach is to develop new graph theoretical infrastructu
res to provide graphical secure channels between nodes in a communication
network of an arbitrary topology.
– Designing distributed algorithms
that can handle various adversarial settings, such as, node crashes and By
zantine attacks. We will mainly provide general compilation schemes that a
re based on exploiting the high-connectivity of the graph. Our key focus w
ill be on the efficiency of the resilient algorithms in terms of the numbe
r of communication rounds.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 201
UID:123se24012024103450
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20230115T133000
DTEND;TZID="Asia/Jerusalem":20230115T143000
DTSTAMP;TZID="Asia/Jerusalem":20230115T133000
FREEBUSY;FBTYPE=BUSY:20230115T133000/20230115T143000
SUMMARY;LANGUAGE=en-US:msc talk by Haitham Fadila about Kernel-based Const
ruction Operators for Boolean Sum and Ruled Geometry at 2023-01-15 13:30:0
0
DESCRIPTION;LANGUAGE=en-US:Boolean sum and ruling are two well-known const
ruction operators for both parametric surfaces and trivariates.
In ma
ny 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 par
ametric geometry so that it has positive Jacobian throughout the domain.
In this work, we focus on cases in which those constructors fail to bu
ild parametric geometries with a positive Jacobian throughout while the fr
eeform input has a kernel point.
We show that in the limit, for high
enough degree raising or enough refinement, our construction scheme must s
ucceed if a kernel exists.
In 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 constr
uction examples are successful.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture: 95299541427 and Taub 301
UID:123se24012024103390
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20230115T143000
DTEND;TZID="Asia/Jerusalem":20230115T153000
DTSTAMP;TZID="Asia/Jerusalem":20230115T143000
FREEBUSY;FBTYPE=BUSY:20230115T143000/20230115T153000
SUMMARY;LANGUAGE=en-US:Coding_Theory_Semina talk by Dor Elimelech (Ben-Gur
ion university about Coding Theory: The Generalized Covering Radius of Co
des at 2023-01-15 14:30:00
DESCRIPTION;LANGUAGE=en-US:The generalized covering radius (GCR) was recen
tly introduced as a fundamental property of linear codes, shown to charact
erize a trade-off between storage amount, access complexity, and latency i
n linear data querying protocols (such as many PIR protocols). In the gene
ral case (where the codes are not necessarily linear), the GCR is used in
order to formulate a higher-order version of the famous combinatorial foot
ball-pool problem. During this talk, we shall discuss the equivalent defin
itions and basic properties of the GCR and survey the recent progress in t
he study of generalized covering codes.
Dor Elimelech received his B.
Sc. in mathematics and his B.Sc. in electrical engineering in 2018 from Be
n-Gurion University of the Negev, Israel; his M.Sc. degree in electrical e
ngineering in 2020 from Ben-Gurion University of the Negev (summa cum laud
e). In 2020 he started his Ph.D. in electrical engineering, also at Ben-Gu
rion University, supervised by Prof. Moshe Schwartz and Prof. Tom Meyerovi
tch. His research interests include coding theory, probability, and dynami
cal systems.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 601
UID:123se24012024103560
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20230115T150000
DTEND;TZID="Asia/Jerusalem":20230115T160000
DTSTAMP;TZID="Asia/Jerusalem":20230115T150000
FREEBUSY;FBTYPE=BUSY:20230115T150000/20230115T160000
SUMMARY;LANGUAGE=en-US:phd talk by Avi Kaplan about Computational Complex
ity under Communication Constraints at 2023-01-15 15:00:00
DESCRIPTION;LANGUAGE=en-US:Can preprocessing help reduce computational cos
ts? We study this question in the context of communication complexity, foc
using on a simple "simultaneous messages" setting in which computationally
unbounded Alice and Bob each send a single message to a computationally b
ounded Carol.
A big part of our work concentrates on the task of comp
uting the inner product function modulo 2 by a polynomial-sized bounded-de
pth Boolean circuit. Without preprocessing this task was shown to be impos
sible, and we show that this is still not possible even if we allow prepro
cessing limited to sublinear stretch of the inputs. Another part of our wo
rk goes beyond inner product and bounded-depth circuits, and explores othe
r computational problems and constraints on Carol in the simultaneous mess
ages framework.
The above question turns out to be closely related to
another question, which is independently motivated by cryptographic appli
cations. Suppose that two distributions X and Y are k-indistinguishable, i
n the sense that their projections to any k coordinates are identically di
stributed. Can some constant-depth circuit distinguish between X and Y? A
celebrated theorem by Braverman implies a negative answer when X is unifor
m, whereas a work of Bogdanov et al. shows that this is not the case in ge
neral. We initiate a systematic study of this question for natural classes
of "simple" distributions, including ones that arise in cryptographic app
lications, obtaining positive results, negative results, and barriers.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture: 95627681547 and Taub 301
UID:123se24012024103430
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20230115T150000
DTEND;TZID="Asia/Jerusalem":20230115T160000
DTSTAMP;TZID="Asia/Jerusalem":20230115T150000
FREEBUSY;FBTYPE=BUSY:20230115T150000/20230115T160000
SUMMARY;LANGUAGE=en-US:phd talk by David Naori about New Models and Improv
ed Bounds for Online Optimization at 2023-01-15 15:00:00
DESCRIPTION;LANGUAGE=en-US: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 a
dversarial input to the online player ahead of time.
The online playe
r competes with the expected optimal value on the part of the input that a
rrives online. Our model bridges between existing online stochastic models
(e.g., items are drawn i.i.d. from a distribution) and the online worst-c
ase model. We also extend in a similar manner (by revealing a sample) the
online random-order model.
We study the secretary problem and online
matching problems in our new models. We also study the online facility loc
ation problem and obtain improved bounds in the standard random-order mode
l.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture: 91014628593 and Taub 401
UID:123se24012024103490
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20230117T113000
DTEND;TZID="Asia/Jerusalem":20230117T123000
DTSTAMP;TZID="Asia/Jerusalem":20230117T113000
FREEBUSY;FBTYPE=BUSY:20230117T113000/20230117T123000
SUMMARY;LANGUAGE=en-US:pixel-club talk by Mark Sheinin (Carnegie Mellon) a
bout Pixel Club: Computational Imaging for Enabling Vision Beyond Human Pe
rception at 2023-01-17 11:30:00
DESCRIPTION;LANGUAGE=en-US:From minute surface vibrations to very fast-occ
urring events, the world is rich with phenomena humans cannot perceive. Li
kewise, 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 thi
s talk, I will show that we can capture these hidden phenomena by creative
ly building novel vision systems composed of common off-the-shelf componen
ts (i.e., cameras and optics) coupled with cutting-edge algorithms.
S
pecifically, I will cover three projects using computational imaging to se
nse 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 wil
l describe Diffraction Line Imaging, a novel imaging principle that exploi
ts 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 appr
oach for sensing minute high-frequency surface vibrations (up to 63kHz) fo
r multiple scene sources simultaneously, using "slow" sensors rated for on
ly 130Hz. Applications include capturing vibration caused by audio sources
(e.g., speakers, human voice, and musical instruments) and localizing vib
ration sources (e.g., the position of a knock on the door).
Bio:
Mark Sheinin is a Post-doctoral Research Associate at Carnegie Mellon Univ
ersity's Robotic Institute at the Illumination and Imaging Laboratory. He
received his Ph.D. in Electrical Engineering from the Technion - Israel In
stitue 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 202
2. He received the Porat Award for Outstanding Graduate Students, the Jaco
bs-Qualcomm Fellowship in 2017, and the Jacobs Distinguished Publication A
ward in 2018. His research interests include computational photography and
computer vision.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Room 1003, EE Meyer Building
UID:123se24012024103630
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20230118T123000
DTEND;TZID="Asia/Jerusalem":20230118T133000
DTSTAMP;TZID="Asia/Jerusalem":20230118T123000
FREEBUSY;FBTYPE=BUSY:20230118T123000/20230118T133000
SUMMARY;LANGUAGE=en-US:Theory Seminar talk by Tal Herman (Weizmann Institu
te of Science) about Theory Seminar: Verifying The unseen: Interactive pro
ofs for Label-Invariant distribution Properties at 2023-01-18 12:30:00
DESCRIPTION;LANGUAGE=en-US:Given i.i.d. samples from an unknown distributi
on over a large domain [N], approximating several basic quantities, includ
ing the distribution’s support size, its entropy, and its distance from th
e uniform distribution, requires (NlogN) samples [Valiant and Valiant, STO
C 2011].
Suppose, however, that we can interact with a powerful but u
ntrusted prover, who knows the entire distribution (or a good approximatio
n of it). Can we use such a prover to approximate (or rather, to approxima
tely {\em verify}) such statistical quantities more efficiently? We show t
hat this is indeed the case: the support size, the entropy, and the distan
ce from the uniform distribution, can all be approximately verified via a
2-message interactive proof, where the communication complexity, the verif
ier’s running time, and the sample complexity are O(N) . For all these qua
ntities, the sample complexity is tight up to \polylogN factors (for any i
nteractive proof, regardless of its communication complexity or verificati
on time).
More generally, we give a tolerant interactive proof system
with the above sample and communication complexities for verifying a dist
ribution’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) , u
nder a mild assumption about the complexity of deciding, given a compact r
epresentation 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:123se24012024103580
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20230118T123000
DTEND;TZID="Asia/Jerusalem":20230118T143000
DTSTAMP;TZID="Asia/Jerusalem":20230118T123000
FREEBUSY;FBTYPE=BUSY:20230118T123000/20230118T143000
SUMMARY;LANGUAGE=en-US:CSpecial Event about Recruitment Day by Mobileye a
t 2023-01-18 12:30:00
DESCRIPTION;LANGUAGE=en-US:Mobileye representatives will visit CS to prese
nt the development of the software and algorithms of Mobileye's autonomous
vehicle, the possibilities of employment and life in the company, on Wedn
esday, 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:123se24012024103610
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20230119T103000
DTEND;TZID="Asia/Jerusalem":20230119T113000
DTSTAMP;TZID="Asia/Jerusalem":20230119T103000
FREEBUSY;FBTYPE=BUSY:20230119T103000/20230119T113000
SUMMARY;LANGUAGE=en-US:colloq talk by Guy Tamir (Technology Evangelist, In
tel) about CS Guest Lecture: Open Software for the Parallel, Heterogeneous
Future at 2023-01-19 10:30:00
DESCRIPTION;LANGUAGE=en-US:The race for performance and the variety of spe
cialized workloads drives the industry to build more parallel, heterogenou
s, and distributed computing systems. These systems introduce multiple pro
gramming challenges. This talk will overview the driving forces, world tre
nds, challenges, and emerging solutions. Specifically, we will overview th
e oneAPI Initiative and its components and benefits. We will demonstrate S
YCL's new programming paradigm and more.
Biography: Guy Tamir is a techn
ology evangelist at Intel Software and Advanced Technology group. His main
areas of interest and expertise are Artificial Intelligence, Computer vis
ion, Video processing, and Heterogeneous, multi-accelerator parallel compu
ting. 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://youtub
e.com/playlist?list=PLg-UKERBljNxsCltpcXU_Haz9xQSCN_SB
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture: 97146417324
UID:123se24012024103370
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20230122T103000
DTEND;TZID="Asia/Jerusalem":20230122T113000
DTSTAMP;TZID="Asia/Jerusalem":20230122T103000
FREEBUSY;FBTYPE=BUSY:20230122T103000/20230122T113000
SUMMARY;LANGUAGE=en-US:colloq talk by Michal Moshkovitz (Bosch Center & Te
l-Aviv University) about CS Lecture: Building the Foundations of Explainab
le and Interpretable Machine Learning at 2023-01-22 10:30:00
DESCRIPTION;LANGUAGE=en-US:Machine learning (ML) is integrated into our so
ciety, it is present in the judicial, health, transportation, and financi
al 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.
Despite the importance of explainability and interpretability, their
foundations are missing. Basic questions are left unanswered: How to defin
e explainability and interpretability? Is there a tradeoff between perform
ance and interpretability? How to evaluate the quality of explanation? In
this talk we start answering these questions in the realm of supervised, u
nsupervised, and reinforcement learning.
Bio:
Michal is a research
scientist at Bosch Center for AI and a visiting researcher at Tel-Aviv Un
iversity. Previously, she was a postdoctoral fellow at the Qualcomm Instit
ute of the University of California San Diego and a postdoc at Tel-Aviv Un
iversity 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 ma
thematical foundations of explainable machine learning.
Michal receiv
ed her Ph.D. from the Hebrew University and an MSc from Tel-Aviv Universit
y. During her Ph.D., Michal interned at the Machine Learning for Healthcar
e and Life Sciences group of IBM Research and the Foundations of Machine L
earning group of Google. Michal has been selected as a 2021 EECS MIT Risin
g Star, the recipient of the Anita Borg scholarship from Google and the Ho
ffman scholarship from the Hebrew University.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 601
UID:123se24012024103650
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20230124T123000
DTEND;TZID="Asia/Jerusalem":20230124T143000
DTSTAMP;TZID="Asia/Jerusalem":20230124T123000
FREEBUSY;FBTYPE=BUSY:20230124T123000/20230124T143000
SUMMARY;LANGUAGE=en-US:CSpecial Event about Projects Fair on IoT, Android
, Arduino and Networks at 2023-01-24 12:30:00
DESCRIPTION;LANGUAGE=en-US:You are invited to the CS Taub projects fair fo
r the Winter Semester of 2023, where 30 teams of
undergraduate students will present and demonstrate projects in various f
ields in IoT, Android, Arduino and Networks, developed as part of the fin
al project in the software engineering and communication networks track, m
ost of which were carried out in collaboration with various social associa
tions and organizations, and were intended to make a contribution to the c
ommunity.
The fair will be held on Tuesday,
January 24, 2023, 12:3
span>0-14:30, at the C
S Taub Lobby.
The presenting posters (Heb)
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:CS Taub Lobby
UID:123se24012024103620
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20230125T123000
DTEND;TZID="Asia/Jerusalem":20230125T133000
DTSTAMP;TZID="Asia/Jerusalem":20230125T123000
FREEBUSY;FBTYPE=BUSY:20230125T123000/20230125T133000
SUMMARY;LANGUAGE=en-US:Theory Seminar talk by Michal Wlodarczyk (Ben-Gurio
n University) about Theory Seminar: Hitting Minors, Planarization, and Ker
nelization at 2023-01-25 12:30:00
DESCRIPTION;LANGUAGE=en-US: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 algor
ithmic paradigms: approximation, kernelization, and parameterized algorith
ms. Here, an important special case is the Vertex Planarization problem (r
emove as few vertices as possible to make a given graph planar) – this pro
blem 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 kerneli
zation. This is a joint work with Bart. M. P. Jansen. No prior background
on graph minors or kernelization is required.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 201
UID:123se24012024103590
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20230126T103000
DTEND;TZID="Asia/Jerusalem":20230126T113000
DTSTAMP;TZID="Asia/Jerusalem":20230126T103000
FREEBUSY;FBTYPE=BUSY:20230126T103000/20230126T113000
SUMMARY;LANGUAGE=en-US:CSpecial Talk talk by Tim Mattson (Senior principal
engineer, Intel) about CS Guest Lecture: Software Development in the Sixt
h Epoch of Distributed Computing at 2023-01-26 10:30:00
DESCRIPTION;LANGUAGE=en-US:Amin Vahdat, in a talk that has gone viral, des
cribed the five epochs of distributed computing (https://www.youtube.com/w
atch?v=Am_itCzkaE0). It’s a great talk, but I disagree with him on one ke
y 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.
In this talk I will very b
riefly outline the five Epochs of distributed computing and then shift to
the future and the sixth Epoch. This Epoch emerges when we bring next gene
ration 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 refe
rence in DRAM (Distributed Random Access Memory).
This innovation is
coming in the not-too-distant future. It will fundamentally change how hi
gh-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 ho
pe this talk launches an aggressive, and hopefully interesting, dialog abo
ut software development in the sixth epoch of distributed computing.
Biography: Tim Mattson is a parallel programmer obsessed with every variet
y of science (Ph.D. Chemistry, UCSC, 1985). He is a senior principal engi
neer in Intel’s parallel computing lab. Tim has been with Intel since 199
3 and has worked with brilliant people on great projects including: (1) th
e first TFLOP computer (ASCI Red), (2) MPI, OpenMP and OpenCL, (3) two dif
ferent research processors (Intel's TFLOP chip and the 48 core SCC), (4) D
ata management systems (Polystore systems and Array-based storage engines)
, and (5) the GraphBLAS API for expressing graph algorithms as sparse line
ar algebra. Tim has well over 150 publications including five books on dif
ferent aspects of parallel computing, the latest (Published November 2019)
titled “The OpenMP Common Core: making OpenMP Simple Again”.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture: 94604196201
UID:123se24012024103350
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20230126T110000
DTEND;TZID="Asia/Jerusalem":20230126T120000
DTSTAMP;TZID="Asia/Jerusalem":20230126T110000
FREEBUSY;FBTYPE=BUSY:20230126T110000/20230126T120000
SUMMARY;LANGUAGE=en-US:msc talk by Ariel Larey about Develop Novel Compute
r Vision and Deep Learning Techniques for Digital Pathology at 2023-01-26
11:00:00
DESCRIPTION;LANGUAGE=en-US:The diagnosis and treatment planning of many di
seases, such as cancer and auto-immune conditions, rely on histological sl
ides. In recent years, digital pathology has become more abundant allowing
high-thruput digitization of pathology images and the use of AI to analyz
e and interpret them. Yet, there are still inherent challenges in harnessi
ng AI for pathology that includes coping with features in multiple size sc
ales, the ability to achieve interpretability of the AI results, and biase
d datasets that impede the ability to produce reliable decision systems.
In the first part of the talk, we will present our AI-based decision s
upport system for the diagnosis of Eosinophilic esophagitis (EoE), a chron
ic immune disease that is second only to gastroesophageal reflux disease a
s the leading cause of chronic refractory dysphagia in adults and children
. Diagnostics of EoE rely on counting single immune cells within a huge wh
ole-slide image, a time-consuming process that is prone to errors. Our pla
tform goes beyond recapturing the current manual histological gold standar
d by AI and reveals novel local and spatial biomarkers for EoE diagnosis,
and can be harnessed to the diagnostics of other conditions.
In the s
econd part, we will present a novel approach for generating synthetic sema
ntic masks of histological tissues. Many histology datasets are biased due
to biological factors (many healthy patients and many very sick but not e
nough around the decision threshold) or technical factors (images from a p
articular device or a specific campus). While GAN-based solutions can prod
uce realistic textures, their ability to recapitulate the spatial distribu
tion of features in tissues is limited. One solution is to use image trans
lation conditional networks, however, generating proper conditional masks
of tissues, as input to the network, is still not done successfully. We wi
ll present our approach that can produce realistic semantic masks of vario
us 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 p
athology solutions.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture: 98200430832 and Faculty of Medicine, seminar room 4th floor
UID:123se24012024103570
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20230130T143000
DTEND;TZID="Asia/Jerusalem":20230130T153000
DTSTAMP;TZID="Asia/Jerusalem":20230130T143000
FREEBUSY;FBTYPE=BUSY:20230130T143000/20230130T153000
SUMMARY;LANGUAGE=en-US:msc talk by Eli Gavril about TCAN: Authentication W
ithout Cryptography on a CAN Bus Based on Nodes Location on the Bus at 202
3-01-30 14:30:00
DESCRIPTION;LANGUAGE=en-US:Vehicles possess an extraordinary amount of tec
hnological features that are meant to improve the safety and comfort of th
e driving experience. Those features have become so advanced that many of
the driving aspects are now almost completely automated. Most drivers in t
he world now rely on the computer systems of the vehicle itself in order t
o perform even the most basic tasks, such as steering and parking.
Th
e CAN bus is the main network used for communication between the various s
ystems of the vehicle. As such, it is a major target for attackers who wis
h to break into the car. Indeed, it has been proven that attacks can be pe
rformed on the CAN bus in order to cause physical damage to the vehicle. S
pecifically, 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.
In this t
hesis we present TCAN, an authentication mechanism for messages on the CAN
bus that does not require cryptography. TCAN ensures that the messages ar
e sent by their alleged senders, and are not modified by other parties con
nected 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 pr
opagation on the bus).
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture: 8355062003
UID:123se24012024103600
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20230201T103000
DTEND;TZID="Asia/Jerusalem":20230201T113000
DTSTAMP;TZID="Asia/Jerusalem":20230201T103000
FREEBUSY;FBTYPE=BUSY:20230201T103000/20230201T113000
SUMMARY;LANGUAGE=en-US:colloq talk by Aviad Levis (Computing and Mathemati
cs at Caltech) about CS Lecture: Computational Imaging for Scientific Disc
overy: From Cloud Physics to Black Holes Dynamics at 2023-02-01 10:30:00
DESCRIPTION;LANGUAGE=en-US:Imaging plays a key role in advancing science,
from revealing the internal structure of clouds to providing the first vis
ual evidence of a black hole. While both examples come from different imag
ing systems, they illustrate what can be achieved with modern computationa
l approaches. Computational imaging combines concepts from physics, machin
e learning, and signal processing to reveal hidden structures at the small
est and largest of scales. In this talk, I will highlight how peeling away
layers of the underlying physics leads to a spectrum of algorithms target
ing new scientific discoveries. I will focus on the Event Horizon Telescop
e (EHT), a unique computational camera with the goal of imaging the glowin
g fluid surrounding supermassive black holes. In May of 2022, the EHT coll
aboration revealed the first images of the black hole at the center of our
galaxy: Sagittarius A* (Sgr A*). These images were computationally recons
tructed from measurements taken by synchronized telescopes around the glob
e. While images certainly offer interesting insights, looking toward the f
uture, we are developing new computational algorithms that aim to go beyon
d a 2D image. For example, could we use EHT observations to recover the dy
namic evolution or even the 3D structure? We tackle these challenges by in
tegrating emerging AI concepts with physics models. Our hope is that in th
e not-too-distant future, these new and exciting prospects will enable sci
entific discovery and even provide a glimpse into the very nature of space
-time itself in our galaxy's most extreme environment.
Bio: Aviad L
evis is a postdoctoral scholar in the Department of Computing and Mathemat
ics at Caltech, working with Katie Bouman. Currently, as part of the Event
Horizon Telescope collaboration, his work focuses on developing computati
onal algorithms for imaging black hole dynamics. Prior to that, he receive
d 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 h
as 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 recipi
ent of the Zuckerman and the Viterbi Postdoctoral Fellowships.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 601
UID:123se24012024103700
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20230201T113000
DTEND;TZID="Asia/Jerusalem":20230201T123000
DTSTAMP;TZID="Asia/Jerusalem":20230201T113000
FREEBUSY;FBTYPE=BUSY:20230201T113000/20230201T123000
SUMMARY;LANGUAGE=en-US:msc talk by Matan Yechieli about Low-Latency Blockc
hains with DAG Holography at 2023-02-01 11:30:00
DESCRIPTION;LANGUAGE=en-US:Classical Proof-of-Work blockchains like Bitcoi
n implement a decentralized ledger, where anyone can participate. They agg
regate transactions from system users in blocks and decide each block's po
sition in the ledger. They require the block at each position to accrue vo
tes until the probability of a decision change, due to chance or malice, i
s 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.
We pre
sent Holograph, a blockchain protocol that achieves a latency of seconds u
nder practical bandwidth limitations. To achieve this we introduce a novel
technique called dag (Directed Acyclic Graph) Holography. Holograph parti
cipants form a single physical block dag. Each physical block manifests as
multiple virtual blocks in multiple virtual dags that serve as parallel v
oting structures.By viewing the same physical dag as many virtual ones, li
ke a holographic image viewed from different angles, we obtain more votes
with the same bandwidth.
We analyze Holograph’s latency as a functio
n of overhead bandwidth limit, compared against prior art. To the best of
our knowledge, this is the first such analysis of blockchain protocols. Ou
r simulation shows that Holograph reaches the lowest latency across the te
sted 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-m
agnitude lower, practical overhead throughput.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture: 93583582399 and Taub 401
UID:123se24012024103680
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20230202T150000
DTEND;TZID="Asia/Jerusalem":20230202T160000
DTSTAMP;TZID="Asia/Jerusalem":20230202T150000
FREEBUSY;FBTYPE=BUSY:20230202T150000/20230202T160000
SUMMARY;LANGUAGE=en-US:CSpecial Talk talk by Hagit Attiya (CS, Technion) a
nd Constantin Enea (École Polytechnique/CNRS) about EuroTech: Using Concur
rent Objects in Randomized Programs at 2023-02-02 15:00:00
DESCRIPTION;LANGUAGE=en-US:Atomic concurrent objects, whose operations tak
e place instantaneously, are a powerful technique for designing complex co
ncurrent 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 lineariz
ability does not preserve hyper-properties, which include probabilistic gu
arantees about randomized programs. A more restrictive property, strong li
nearizability, does preserve hyper-properties but it is impossible to achi
eve in many situations.
In particular, we show that there are no stro
ngly linearizable implementations of multi-writer registers or snapshot ob
jects in message-passing systems. On the other hand, we show that a wide c
lass of linearizable implementations, including well-known ones for regist
ers and snapshots, can be modified to approximate the probabilistic guaran
tees of randomized programs when using atomic objects.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Zoom Lecture:
Registration
UID:123se24012024103670
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20230207T103000
DTEND;TZID="Asia/Jerusalem":20230207T113000
DTSTAMP;TZID="Asia/Jerusalem":20230207T103000
FREEBUSY;FBTYPE=BUSY:20230207T103000/20230207T113000
SUMMARY;LANGUAGE=en-US:colloq talk by Nadav Amit (VMware Research) about C
S Lecture: Reexamining Basic OS Memory Management Techniques at 2023-02-07
10:30:00
DESCRIPTION;LANGUAGE=en-US:Despite significant advancements in operating s
ystem memory management, our understanding of the desired behavior of fund
amental techniques introduced decades ago is sometimes incomplete or not w
ell-defined. This can result in correctness issues that might cause the sy
stem to crash or be compromised, as well as missed opportunities for optim
izations. In this talk, I will present two specific examples of this: (1)
the inefficiencies in synchronizing the memory view across different CPU c
ores, and (2) the undefined behavior of the interactions between two commo
n memory management mechanisms, copy-on-write and pinning.
Bio:
N
adav Amit is a senior researcher in VMware Research. He received his PhD i
n 2014 from the Technion, Israel Institute of Technology for his work on a
lleviating of virtualization bottlenecks. He is a recipient of the SPEC Di
stinguished Dissertation Award, IBM Fellowship Award and an honorable ment
ion of the Dennis M. Ritchie Doctoral Dissertation Award. His current rese
arch revolves operating systems and virtualization and focuses on memory m
anagement.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 601
UID:123se24012024103730
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20230208T153000
DTEND;TZID="Asia/Jerusalem":20230208T163000
DTSTAMP;TZID="Asia/Jerusalem":20230208T153000
FREEBUSY;FBTYPE=BUSY:20230208T153000/20230208T163000
SUMMARY;LANGUAGE=en-US:msc talk by Almog Zur about RELAX: Recovering Lazil
y from Failed Execution with
Persistent Memory at 2023-02-08 15:30:00
DESCRIPTION;LANGUAGE=en-US:Recent non-volatile main memory technology (suc
h as Intel’s Optane) gave rise to an abundance of research on building per
sistent data structures, whose content can be recovered after a system cra
sh. While there has been significant progress in making durable data struc
tures efficient, shortening the length of the recovery phase after a crash
(in which data cannot be accessed) has not received much attention. In fa
ct, 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 p
aper we present the RELAX general transformation. RELAX generates durable
data structures that provide the best of both worlds. They provide high pe
rformance with almost zero recovery time, with an overhead that quickly de
scends following a crash event, until the program soon regains maximal per
formance. 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 dat
a structures are fast, and that following a crash, even large data sets wi
th hundreds of millions of nodes become responsive within a few millisecon
ds, whereas other efficient constructions require more than 10 seconds of
unresponsive recovery time.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 601
UID:123se24012024103640
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20230212T113000
DTEND;TZID="Asia/Jerusalem":20230212T123000
DTSTAMP;TZID="Asia/Jerusalem":20230212T113000
FREEBUSY;FBTYPE=BUSY:20230212T113000/20230212T123000
SUMMARY;LANGUAGE=en-US:msc talk by Gilad Chase about Generalized polymorph
isms at 2023-02-12 11:30:00
DESCRIPTION;LANGUAGE=en-US: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 a
ll Boolean inputs $\{ z_{ij} : i \in [n], j \in [m] \}$. This extends char
acterizations 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:123se24012024103710
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20230212T123000
DTEND;TZID="Asia/Jerusalem":20230212T133000
DTSTAMP;TZID="Asia/Jerusalem":20230212T123000
FREEBUSY;FBTYPE=BUSY:20230212T123000/20230212T133000
SUMMARY;LANGUAGE=en-US:phd talk by Aviv A. Rosenberg about One sequence, o
ne structure? Computationally identifying protein structures that defy the
central dogma of biology at 2023-02-12 12:30:00
DESCRIPTION;LANGUAGE=en-US:Proteins fold from a sequence of amino acids, f
orming secondary structures which subsequently fold into a three-dimension
al structure that enables their function. The amino acid sequence is defin
ed 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" dogm
a, established over half a century ago, remains the commonly accepted noti
on, and implies that synonymous coding is inconsequential to protein struc
ture. This talk will present results from three different works, which cha
llenge this dogma through large-scale computational analysis of protein st
ructures.
First, we develop novel methods for computing and compari
ng codon-specific protein backbone angle distributions. We design a non-pa
rametric approach for comparing these bivariate distributions using finite
samples, and identify synonymous codon distributions which are distinguis
hable, with statistical significance, within some secondary structures. Th
is demonstrates, for the first time, an association between synonymous cod
on usage and the final protein structure around the amino acids they trans
late into.
Next, we expand this approach to consider pairs of amino
acids, accounting for the peptide bond which is formed between amino acid
s during translation of the genetic code. To that end, we introduce a tool
for defining local, two amino acid-long sub-secondary structural units. W
e analyze the joint distribution of backbone angles across the peptide bon
d and show that our structural units can more meaningfully represent backb
one conformations than conventional secondary structure.
Finally, b
uilding 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 wi
th an identical sequence and potential interaction network. We show that s
uch conformational differences are stable under molecular dynamics simulat
ions, and that they are not predicted by AlphaFold, a state-of-the-art str
ucture prediction model which relies only on the amino acid sequence. Our
data-driven approach provides biologists with invaluable dogma-defying exa
mples, 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:123se24012024103660
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20230215T123000
DTEND;TZID="Asia/Jerusalem":20230215T133000
DTSTAMP;TZID="Asia/Jerusalem":20230215T123000
FREEBUSY;FBTYPE=BUSY:20230215T123000/20230215T133000
SUMMARY;LANGUAGE=en-US:Theory Seminar talk by Seth pettie (University of M
ichigan) about Theory seminar: Algorithms Should Have Bullshit Detectors!
(or Polynomial Time Byzantine Agreement with Optimal Resilience) at 2023-0
2-15 12:30:00
DESCRIPTION;LANGUAGE=en-US:One thing that distinguishes (theoretical) comp
uter science from other scientific disciplines is its full-throated suppor
t of a fundamentally adversarial view of the universe. Malicious adversari
es, with unbounded computational advantages, attempt to foil our algorithm
s 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 buil
ding a “bullshit detector” for algorithms a good idea?
To illustrate
the power of the Bullshit Detection worldview, we give the first polynomia
l-time protocol for Byzantine Agreement that is resilient to f < n/3 corru
ptions against an omniscient, computationally unbounded adversary in an as
ynchronous 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 pro
tocol in which corrupted parties (who chose the outcomes of their coins ma
liciously) are eventually detected via statistical tests.
We will als
o 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:123se24012024103740
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20230216T113000
DTEND;TZID="Asia/Jerusalem":20230216T123000
DTSTAMP;TZID="Asia/Jerusalem":20230216T113000
FREEBUSY;FBTYPE=BUSY:20230216T113000/20230216T123000
SUMMARY;LANGUAGE=en-US:cggc talk by Rephael wenger (The ohio State Univers
ity) about CGGC Seminar: Intro to Morse Theory, Morse-Smale Complexes, and
Discrete Morse Theory at 2023-02-16 11:30:00
DESCRIPTION;LANGUAGE=en-US:Rephael Wenger is a professor in the computer s
cience and engineering department of The Ohio State university where he wo
rks on geometric modeling, mesh generation, geometric algorithms and scien
tific visualization.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il
LOCATION:Taub 401
UID:123se24012024103750
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20230315T113000
DTEND;TZID="Asia/Jerusalem":20230315T123000
DTSTAMP;TZID="Asia/Jerusalem":20230315T113000
FREEBUSY;FBTYPE=BUSY:20230315T113000/20230315T123000
SUMMARY;LANGUAGE=en-US:msc talk by Yaron Hay about A Coloring-based Approa
ch for Concurrent Execution of Transactions and Smart Contracts in Active
Replication and Blockchain Systems at 2023-03-15 11:30:00
DESCRIPTION;LANGUAGE=en-US:Blockchain Networks, especially those with Smar
t Contracts, are well-known examples of Active Replication Systems. Active
Replication Services are available thanks to a group of servers called re
plicas that handle client requests. Each server maintains a local copy of
the global state of the service, and all servers update their local copy a
t synchronized incremental steps. At the i-th step, all servers receive th
e *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.
The combination of a consistent global orderi
ng protocol and having all transactions are inherently deterministic guara
ntees that all replicas will eventually have the same exact state (for eac
h step taken). It is imperative for replicated services. However, this com
es at a loss in performance because of the restriction to executing only o
ne transaction at any given time. Sequential execution prevents us from ut
ilizing the capabilities of multicore processors by processing multiple tr
ansactions in parallel.
In this lecture, we will define formal models t
hat allow for consistent concurrent execution while maintaining determinis
tic results at all replicas, and discuss ways to maximize concurrency.
We'll explore the connection between Graph Vertex Coloring and minimizing
overall latency. Then describe practical approaches to solving this proble
m in real-world applications, with experimental evaluation for two popular
benchmarking frameworks.
Lastly, we'll describe how to implement concu
rrent 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:123se24012024103760
END:VEVENT
BEGIN:VEVENT
DTSTART;TZID="Asia/Jerusalem":20230315T113000
DTEND;TZID="Asia/Jerusalem":20230315T123000
DTSTAMP;TZID="Asia/Jerusalem":20230315T113000
FREEBUSY;FBTYPE=BUSY:20230315T113000/20230315T123000
SUMMARY;LANGUAGE=en-US:phd talk by Shir Cohen about Distributed Services U
nder Attack at 2023-03-15 11:30:00
DESCRIPTION;LANGUAGE=en-US:For my PhD thesis seminar, I will be presenting
two of my works related to the security and reliability of distributed se
rvices in the face of Byzantine attacks. In the first work “Not a COINcide
nce: Sub-Quadratic Asynchronous Byzantine Agreement WHP” (DISC’20), I pres
ent a solution for binary Byzantine Agreement (BA) in asynchronous systems
, using a shared coin algorithm based on a VRF and VRF-based committee sam
pling. 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 barrie
r for asynchronous Byzantine Agreement. This work is then used to solve th
e multivalued version of the BA problem.
In the second work “Tame the
Wild with Byzantine Linearizability: Reliable Broadcast, Snapshots, and A
sset Transfer” (DISC’21), I introduce the concept of Byzantine linearizabi
lity and study Byzantine-tolerant emulations of various objects from regis
ters, including reliable broadcast, atomic snapshot, and asset transfer. T
his work proves that there is an f-resilient implementation of such object
s from registers with n processes for f
The prize-w inning competition will last about 30 hours, registration is open and the number of places is limited. ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il LOCATION:CS Taub Building UID:123se24012024103800 END:VEVENT BEGIN:VEVENT DTSTART;TZID="Asia/Jerusalem":20230402T183000 DTEND;TZID="Asia/Jerusalem":20230402T203000 DTSTAMP;TZID="Asia/Jerusalem":20230402T183000 FREEBUSY;FBTYPE=BUSY:20230402T183000/20230402T203000 SUMMARY;LANGUAGE=en-US:CSpecial Event about Practical Workshop by CYE at 2 023-04-02 18:30:00 DESCRIPTION;LANGUAGE=en-US:You are invited to a practical testing workshop by CYE, supervised byTal Sihonov, director of the development group at th e 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. Participating in the workshop requires prerequi site of completing the Introduction to Ssystems Programming course or prac tical experience from a previous job of writing code in Python. Please pre-register - the number of places is limited! ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il LOCATION:Room 337 taub bld. UID:123se24012024103880 END:VEVENT BEGIN:VEVENT DTSTART;TZID="Asia/Jerusalem":20230403T100000 DTEND;TZID="Asia/Jerusalem":20230403T110000 DTSTAMP;TZID="Asia/Jerusalem":20230403T100000 FREEBUSY;FBTYPE=BUSY:20230403T100000/20230403T110000 SUMMARY;LANGUAGE=en-US:phd talk by Gal Yehuda about Connections between Ma chine Learning and Theoretical Computer Science at 2023-04-03 10:00:00 DESCRIPTION;LANGUAGE=en-US:We present connections between machine learning and theoretical computer science. In particular: hardness of data-set gen eration for deep learning problems and connections between randomness and computation in deep learning. ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il LOCATION:Taub 601 UID:123se24012024103900 END:VEVENT BEGIN:VEVENT DTSTART;TZID="Asia/Jerusalem":20230403T113000 DTEND;TZID="Asia/Jerusalem":20230403T123000 DTSTAMP;TZID="Asia/Jerusalem":20230403T113000 FREEBUSY;FBTYPE=BUSY:20230403T113000/20230403T123000 SUMMARY;LANGUAGE=en-US:pixel-club talk by Rebecca Willet (University of Ch icago) about Pixel Club: Machine Learning and Data Assimilation in the Nat ural Sciences and Engineering at 2023-04-03 11:30:00 DESCRIPTION;LANGUAGE=en-US:The potential for machine learning to revolutio nize scientific and engineering research is immense, but its transformativ e power cannot be fully harnessed through the use of off-the-shelf tools a lone. To unlock this potential, novel methods are needed to integrate phys ical 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 ch allenges in the context of inverse problems, data assimilation, and reduce d order modeling. By leveraging ideas from statistics, optimization, scien tific computing, and signal processing, we can develop new and more effect ive machine learning methods that improve predictive accuracy and computat ional efficiency in the natural sciences. Short Bio: Professor of Sta tistics and Computer Science & Director of AI at the Data Science Institut e, with a courtesy appointment at the Toyota Technological Institute at Ch icago. Faculty lead of AI+Science Postdoctoral Fellow program. Prof. W illett completed her Ph.D. in Electrical and Computer Engineering at Rice University in 2005 and was an Assistant then tenured Associate Professor o f Electrical and Computer Engineering at Duke University from 2005 to 2013 . She was an Associate Professor of Electrical and Computer Engineering, H arvey D. Spangler Faculty Scholar, and Fellow of the Wisconsin Institutes for Discovery at the University of Wisconsin-Madison from 2013 to 2018. Wi llett has also held visiting researcher or faculty positions at the Univer sity of Nice in 2015, the Institute for Pure and Applied Mathematics at UC LA in 2004, the University of Wisconsin-Madison 2003-2005, the French Nati onal Institute for Research in Computer Science and Control (INRIA) in 200 3, and the Applied Science Research and Development Laboratory at GE Healt hcare in 2002. ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il LOCATION:Taub 012 (Learning Center Auditorium) UID:123se24012024103940 END:VEVENT BEGIN:VEVENT DTSTART;TZID="Asia/Jerusalem":20230403T173000 DTEND;TZID="Asia/Jerusalem":20230403T193000 DTSTAMP;TZID="Asia/Jerusalem":20230403T173000 FREEBUSY;FBTYPE=BUSY:20230403T173000/20230403T193000 SUMMARY;LANGUAGE=en-US:CSpecial Event about Meeting on: "Democracy, Hi-tec h and the Future Generation" at 2023-04-03 17:30:00 DESCRIPTION;LANGUAGE=en-US:You are invited to a meeting with hi-tech execu tives on "Democracy, Hi-tech and the Future Generation", with the particip ation of: Dr. Kira Radinsky Nadir Izrael Dr. Orna Berry Molly Aden Yo ram Yaacovi Ori Hadomi Dr. YonathanYaniv Maor Farid Avner Rothschild On Monday, April 3, 2023, 17:30, in Taub 1. Please register in advance ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il LOCATION:CS Taub Build. Auditorium 1 UID:123se24012024103950 END:VEVENT BEGIN:VEVENT DTSTART;TZID="Asia/Jerusalem":20230404T143000 DTEND;TZID="Asia/Jerusalem":20230404T153000 DTSTAMP;TZID="Asia/Jerusalem":20230404T143000 FREEBUSY;FBTYPE=BUSY:20230404T143000/20230404T153000 SUMMARY;LANGUAGE=en-US:msc talk by Ella Sheory about Exploring Advanced Ca che Algorithms for the TLB at 2023-04-04 14:30:00 DESCRIPTION;LANGUAGE=en-US:The translation lookaside buffer (``TLB’’) is a small cache that accelerates virtual to physical address translation, whi ch processors typically manage with variants of the least recently used (` `LRU’’) algorithm. Although LRU is simple, it is suboptimal for some workl oads. Our analysis shows that if the processor uses the optimal---but impr actical---Belady algorithm instead of LRU, runtime improves by up to 15% ( and 5% on average) over LRU in single-thread (``ST’’) mode. Runtime furthe r improves by up to 23% (yet still 5% on average) in simultaneous multithr eading (``SMT’’) mode, where the TLB is competitively shared between two t hreads. Given this background, we observe that while past research deve loped various practical caching algorithms to outperform LRU, such researc h was typically evaluated in the context of regular data caches rather tha n the TLB. Our goal in this study is therefore to investigate whether adva nced practical caching algorithms improve TLB performance and, if so, to w hat extent they can approach Belady performance. To this end, we classi fy existing caching algorithms into three groups based on their additional storage requirements: small (less than 10% of the LRU-managed TLB), mediu m (between 10% and 50%), and high (more than 50%). From each group, we sel ect a representative algorithm. We find that the representatives of the sm all and medium groups improve performance by only 1%, on average (and no m ore than 11%). We also find that the third algorithm's complexity and soph istication are unjustified, as we can achieve the aforementioned optimal B elady improvement by handing the associated additional storage to the simp ler LRU baseline. We thus conclude that no existing practical caching algo rithm can meaningfully improve LRU TLB performance. ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il LOCATION:Taub 301 UID:123se24012024103910 END:VEVENT BEGIN:VEVENT DTSTART;TZID="Asia/Jerusalem":20230404T150000 DTEND;TZID="Asia/Jerusalem":20230404T160000 DTSTAMP;TZID="Asia/Jerusalem":20230404T150000 FREEBUSY;FBTYPE=BUSY:20230404T150000/20230404T160000 SUMMARY;LANGUAGE=en-US:Coding_Theory_Semina talk by Prof. Antonia Wachter- Zeh (University of Munich) about Coding Theory: Interleaved (Rank-Metric) Codes for Cryptography at 2023-04-04 15:00:00 DESCRIPTION;LANGUAGE=en-US:Public-key cryptography is the foundation for e stablishing secure communication between multiple parties. Traditional pub lic-key algorithms such as RSA are based on the hardness of factoring larg e numbers or the discrete logarithm problem, but can be attacked in polyno mial time once a capable quantum computer exists. Code-based public-key cr yptosystems 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. In this talk, we first gi ve an overview of interleaving and decoding algorithms in the Hamming and rank metric and then present different approaches to define code-based cry ptographic schemes. Antonia 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 obtai ned her Ph.D. degree in 2013 from Ulm University and from Universite de Re nnes 1, Rennes, France. From 2013 to 2016, she was a postdoctoral research er at the Technion—Israel Institute of Technology, Haifa, Israel, and from 2016 to 2020 a Tenure Track Assistant Professor at TUM. She is a recipien t 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 informa tion theory and their application to storage, communications, privacy, sec urity and machine learning. ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il LOCATION:Taub 601 UID:123se24012024103930 END:VEVENT BEGIN:VEVENT DTSTART;TZID="Asia/Jerusalem":20230419T123000 DTEND;TZID="Asia/Jerusalem":20230419T143000 DTSTAMP;TZID="Asia/Jerusalem":20230419T123000 FREEBUSY;FBTYPE=BUSY:20230419T123000/20230419T143000 SUMMARY;LANGUAGE=en-US:CSpecial Event about CS Open Day for Graduate Studi es at 2023-04-19 12:30:00 DESCRIPTION;LANGUAGE=en-US:Technion CS open day 2023 invites outstanding u ndergraduates from all universities to learn about the Computer Science De partment and register for Winter Semester 2023-24. The event will be he ld on Wednesday, April 19, 2022. between 12:30-14:00, room 337, Taub Build ing for Computer Science, Technion. The program will include review on curriculum, research and life at the Technion CS Department: - CS Dean, P rof. Danny Raz - Vice Dean, Prof. Gill Barequet - Dr. Liane Levy-Eitan, Director Research Group at Amazon - Mr. Dean Zadok (Ph.D. students) - Q uestions and answers For more information please contact Graduate Studi es Coordinator L imor Gindin< Attendance at the open day requires pre-registration. ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il LOCATION:Room 337 taub bld. UID:123se24012024103920 END:VEVENT BEGIN:VEVENT DTSTART;TZID="Asia/Jerusalem":20230419T190000 DTEND;TZID="Asia/Jerusalem":20230419T210000 DTSTAMP;TZID="Asia/Jerusalem":20230419T190000 FREEBUSY;FBTYPE=BUSY:20230419T190000/20230419T210000 SUMMARY;LANGUAGE=en-US:CSpecial Event about NVIDIA AI Lecture at 2023-04-1 9 19:00:00 DESCRIPTION;LANGUAGE=en-US:You are invited to a joint lecture by the CS an d EE on behalf of NVIDIA that will review AI and NVIDIA technologies, as w ell as the latest developments in the field of large language models, incl uding effective tools for running large models, on Wednesday April 19, 202 3 at 19:00, at the Faculty of Engineering Electrical, Auditorium 1003, Mey er Building. Please pre-register ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il LOCATION:Room 1003, EE Meyer Building UID:123se24012024103990 END:VEVENT BEGIN:VEVENT DTSTART;TZID="Asia/Jerusalem":20230420T130000 DTEND;TZID="Asia/Jerusalem":20230420T140000 DTSTAMP;TZID="Asia/Jerusalem":20230420T130000 FREEBUSY;FBTYPE=BUSY:20230420T130000/20230420T140000 SUMMARY;LANGUAGE=en-US:CSpecial Talk talk by Adam Kalai (Microsoft Researc h New England) about CS Lecture: The Power of Intelligent Language Models at 2023-04-20 13:00:00 DESCRIPTION;LANGUAGE=en-US: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 sever al interesting applications of these models. First, they can be used to en umerate human stereotypes and discriminatory biases, suggesting that they must be used carefully. Second, they can be used to generate and solve the ir 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 participant s in classic behavioral economic and psychology experiments, such as the u ltimatum 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. Bio: Adam Tauman Kalai is a Senior Principal Researcher at Microsoft Re search 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 Carnegi e 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 inter disciplinary initiative to understand the communication of sperm whales. H e has co-chaired AI and crowdsourcing conferences including the COLT (the Conference on Learning Theory), the HCOMP (the Conference on Human Computa tion) and NEML. His honors include the Majulook prize, multiple best paper awards, an NSF CAREER award, and an Alfred P. Sloan fellowship. ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il LOCATION:Taub 601 UID:123se24012024103980 END:VEVENT BEGIN:VEVENT DTSTART;TZID="Asia/Jerusalem":20230423T143000 DTEND;TZID="Asia/Jerusalem":20230423T153000 DTSTAMP;TZID="Asia/Jerusalem":20230423T143000 FREEBUSY;FBTYPE=BUSY:20230423T143000/20230423T153000 SUMMARY;LANGUAGE=en-US:Coding_Theory_Semina talk by Prof. Uzi Pereg about Coding Theory: The Multiple-Access Channel with Entangled Transmitters at 2023-04-23 14:30:00 DESCRIPTION;LANGUAGE=en-US:Quantum communication has seen rapid developmen t in the last decade, in both practice and theory. Recently, there is a gr owing interest in how quantum entanglement can assist classical networks, i.e., non-quantum communication systems. In particular, there are known ex amples of classical multi-user channels such that the sum rate with entang led transmitters is strictly higher than the best achievable sum rate with out such resources. The present work studies a two-user classical mult iple-access channel (MAC) with entanglement resources shared between the t ransmitters a priori. We determine the capacity region for the general MAC with entangled transmitters, and show that the previous results can be ob tained as a special case. We also point out the following change of behavi or. Without entanglement resources, Dueck (1978) showed that the relaxatio n of a message-average error criterion can lead to strictly higher achieva ble rates, when compared with a maximal error criterion. Here, however, we show that the capacity region with entangled transmitters is the same, wh ether we consider a message-average or a maximal error criterion. Uzi Pereg is an assistant professor at the Viterbi Faculty of Electrical and C omputer Engineering (ECE) and the Hellen Diller Quantum Center in the Tech nion - Israel Institute of Technology. He was a postdoc at the Institute f or 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 Sc ience and Technology Postdoc Fellowship of the Israel Council for Higher E ducation (CHE), the Seed Funding Grant of the Munich Center for Quantum Sc ience and Technology (MCQST), the Chaya Career Advancement Chair, and the VATAT Fellowship for Junior Faculty Members in Quantum Science and Technol ogy. ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il LOCATION:Taub 601 UID:123se24012024103970 END:VEVENT BEGIN:VEVENT DTSTART;TZID="Asia/Jerusalem":20230423T163000 DTEND;TZID="Asia/Jerusalem":20230423T173000 DTSTAMP;TZID="Asia/Jerusalem":20230423T163000 FREEBUSY;FBTYPE=BUSY:20230423T163000/20230423T173000 SUMMARY;LANGUAGE=en-US:phd talk by Sagi Marcovich about Balanced de Bruijn Sequences at 2023-04-23 16:30:00 DESCRIPTION;LANGUAGE=en-US:Balanced sequences and balanced codes have attr acted a lot of research in the last seventy years due to their diverse app lications in information theory as well as other areas of computer science and engineering. There have been some methods to classify balanced sequen ces. This work suggests two new different hierarchies to classify these se quences. 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 proper ty 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 enc oding and decoding algorithms are provided in this work. ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il LOCATION:Zoom Lecture: 92840391109 UID:123se24012024103830 END:VEVENT BEGIN:VEVENT DTSTART;TZID="Asia/Jerusalem":20230423T173000 DTEND;TZID="Asia/Jerusalem":20230423T193000 DTSTAMP;TZID="Asia/Jerusalem":20230423T173000 FREEBUSY;FBTYPE=BUSY:20230423T173000/20230423T193000 SUMMARY;LANGUAGE=en-US:CSpecial Event about Designated Meeting for Graduat e Students: Research Career in Industry at 2023-04-23 17:30:00 DESCRIPTION;LANGUAGE=en-US:You are invited to a designated meeting for gra duate students, with a panel that will deal with research careers in indus try: What does research in industry look like? What is the admission proce ss for research positions? What are the types of jobs available and career paths? Featuring: Dr. Liane Levy-Eitan, Research Group Director, Amazo n Amichai Shulman, cyber entrepreneur and investor Dr. Rachel Tzoref-Bri ll, Senior Researcher, IBM Research Laboratory The meeting will take pl ace on Sunday, May 23, 2023, 17:30, at the Grads Club in Taub (2nd floor). Please pre-register. ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il LOCATION:CS Grads Club, Floor 2, CS Taub Building UID:123se24012024103960 END:VEVENT BEGIN:VEVENT DTSTART;TZID="Asia/Jerusalem":20230427T093000 DTEND;TZID="Asia/Jerusalem":20230427T103000 DTSTAMP;TZID="Asia/Jerusalem":20230427T093000 FREEBUSY;FBTYPE=BUSY:20230427T093000/20230427T103000 SUMMARY;LANGUAGE=en-US:msc talk by Dave Makhervaks about Clinical Contradi ction Detection at 2023-04-27 09:30:00 DESCRIPTION;LANGUAGE=en-US:Detecting contradictions in text is essential i n determining the validity of the literature and sources that we consume. Medical corpora are riddled with conflicting statements. This is due to th e large throughput of new studies and the difficulty in replicating experi ments, such as clinical trials. Detecting contradictions in this domain is hard since it requires clinical expertise. In this work, we present a dis tant supervision approach that leverages a medical ontology to build a see d 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 represen t potential medical contradiction. The dataset is used to weakly-supervise state-of-the-art deep learning models showing significant empirical impro vements across multiple medical contradiction datasets. ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il LOCATION:Zoom Lecture: 99219466853 and Taub 601 UID:123se24012024104000 END:VEVENT BEGIN:VEVENT DTSTART;TZID="Asia/Jerusalem":20230502T103000 DTEND;TZID="Asia/Jerusalem":20230502T113000 DTSTAMP;TZID="Asia/Jerusalem":20230502T103000 FREEBUSY;FBTYPE=BUSY:20230502T103000/20230502T113000 SUMMARY;LANGUAGE=en-US:msc talk by Majd Khoury about On Distributed Comput ation of the Minimum Triangle Edge Transversal at 2023-05-02 10:30:00 DESCRIPTION;LANGUAGE=en-US:In this work, we study the complexity of comput ing the distance of a graph from being triangle-free in distributed settin gs, 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:123se24012024104060 END:VEVENT BEGIN:VEVENT DTSTART;TZID="Asia/Jerusalem":20230502T113000 DTEND;TZID="Asia/Jerusalem":20230502T123000 DTSTAMP;TZID="Asia/Jerusalem":20230502T113000 FREEBUSY;FBTYPE=BUSY:20230502T113000/20230502T123000 SUMMARY;LANGUAGE=en-US:phd talk by Gregory Vaksman about Modern Learning T echnics for Image and Video Denoising Via Patch Matching at 2023-05-02 11: 30:00 DESCRIPTION;LANGUAGE=en-US:Image and video denoising has been an area of r esearch interest for decades. This talk will present three novel methods t hat take the denoising field a step forward. All the proposed methods in t his work strongly rely on exploiting non-local self-similarity using patch matching. The first method, termed LIDIA [1], has two contributions. First, we propose a low-weight architecture that achieves near state-of-th e-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 pe rformance while addressing novel visual content that deviates from the tra ining data. The next method, termed PaCNet [2], is inspired by LIDIA, proposing a novel algorithm for video denoising. As in LIDIA, PaCNet relie s 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. O ur algorithm achieves state-of-the-art performance, surpassing the competi tors on average by about 0.7 dB. The third method [3], inspired by bot h LIDIA and PaCNet, proposes a novel self-supervised training technique su itable for the removal of unknown correlated noise. The proposed approach neither requires knowledge of the noise model nor access to ground-truth t argets. We assume that the noise is additive, zero mean, but not necessari ly Gaussian, and one that could be short-range spatially correlated. Examp les of such noise could be Gaussian correlated noise, shot noise passed th rough a linear space-invariant system, or real image noise in digital came ras. We demonstrate superior denoising performance compared to leading alt ernative self-supervised denoising methods. [1] G. Vaksman, M. Elad, an d P. Milanfar. Lidia: Lightweight learned image denoising with instance ad aptation. 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognit ion Workshops (CVPRW), pages 2220-2229, 2020. [2] G. Vaksman, M. Elad, an d P. Milanfar. Patch craft: Video denoising by deep modeling and patch mat ching. 2021 IEEE/CVF International Conference on Computer Vision (ICCV), p ages 2137-2146, 2021. [3] G. Vaksman and M. Elad. Patch-Craft Self-Superv ised Training for Correlated Image Denoising. To appear in the Conference on Computer Vision and Pattern Recognition (CVPR), 2023. ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il LOCATION:Taub 401 UID:123se24012024104030 END:VEVENT BEGIN:VEVENT DTSTART;TZID="Asia/Jerusalem":20230507T143000 DTEND;TZID="Asia/Jerusalem":20230507T153000 DTSTAMP;TZID="Asia/Jerusalem":20230507T143000 FREEBUSY;FBTYPE=BUSY:20230507T143000/20230507T153000 SUMMARY;LANGUAGE=en-US:msc talk by Avital Boruchovsky about DNA-Correcting Codes: End-to-end Correction in DNA Storage Systems at 2023-05-07 14:30:0 0 DESCRIPTION;LANGUAGE=en-US:Existing storage technologies cannot keep up wi th the modern data explosion. There is a growing need to find alternatives for the current solutions for storing data. Storage systems based DNA, se ems 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). A typical DNA storage system consists of three important components. T he first is the DNA synthesis which produces the oligonucleotides, also ca lled strands, that encode the data. The second part is a storage container with compartments which stores the DNA strands, however without order. Fi nally, to retrieve the data, the DNA is accessed using next-generation seq uencing, which results in several noisy copies, called reads. The retrieva l of the input information, is usually done by three steps as well. The fi rst step is to partition all the reads into clusters such that the reads a t each cluster are all copies of the same information strand. The second s tep is to apply a reconstruction algorithm on every cluster in order to re trieve an approximation of the original input strands. In the last step an Error Correcting Code is used in order to correct the remaining errors an d to retrieve the user’s information. This work presents a new solution to DNA storage that integrates all three steps of retrieval, namely clust ering, reconstruction, and error correction. DNA-correcting codes are pres ented as a unique solution to the problem of ensuring that the output of t he storage system is unique for any valid set of input strands. To this en d, 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 con structions of DNA-correcting codes. ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il LOCATION:Taub 601 UID:123se24012024104040 END:VEVENT BEGIN:VEVENT DTSTART;TZID="Asia/Jerusalem":20230509T113000 DTEND;TZID="Asia/Jerusalem":20230509T123000 DTSTAMP;TZID="Asia/Jerusalem":20230509T113000 FREEBUSY;FBTYPE=BUSY:20230509T113000/20230509T123000 SUMMARY;LANGUAGE=en-US:phd talk by Tomer Weiss about Deep Learning Approac hes for Inverse Problems in Computational Imaging and Chemistry at 2023-05 -09 11:30:00 DESCRIPTION;LANGUAGE=en-US:In this talk, I will present two chapters from my Ph.D. thesis. The core of my research focuses on methods that utilize t he power of modern neural networks not only for their conventional tasks s uch as prediction or reconstruction, but rather use the information they “ learned” (usually in the forms of their gradients) in order to optimize so me end-task, draw insight from the data, or even guide a generative model. The first part of the talk is dedicated to computational imaging and s hows how to apply joint optimization of the forward and inverse models to improve the end performance. We demonstrate these methods on three differe nt tasks in the fields of Magnetic Resonance Imaging (MRI) and Multiple In put Multiple Output (MIMO) radar imaging. In the second part, we show a novel method for molecular inverse design that utilizes the power of neur al networks in order to propose molecules with desired properties. We deve loped a guided diffusion model that uses the gradients of a pre-trained pr ediction 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:123se24012024104090 END:VEVENT BEGIN:VEVENT DTSTART;TZID="Asia/Jerusalem":20230509T143000 DTEND;TZID="Asia/Jerusalem":20230509T153000 DTSTAMP;TZID="Asia/Jerusalem":20230509T143000 FREEBUSY;FBTYPE=BUSY:20230509T143000/20230509T153000 SUMMARY;LANGUAGE=en-US:colloq talk by Dan Halperin (Tel Aviv University) a bout CS Colloquia: From Snapping Fixtures to Multi-robot Coordination: Ge ometry at the Service of Robotics at 2023-05-09 14:30:00 DESCRIPTION;LANGUAGE=en-US:Robots sense, move and act in the physical worl d. It is therefore natural that understanding the geometry of the problem at hand is often key to devising an effective robotic solution. I will rev iew several problems in robotics and automation in whose solution geometry plays a major role. These include designing optimized 3D printable fixtur es, object rearrangement by robot arm manipulators, and efficient coordina tion of the motion of large teams of robots. As we shall see, exploiting g eometric structure can, among other benefits, lead to reducing the dimensi onality of the underlying search space and in turn to efficient solutions. Short bio: Dan Halperin received his Ph.D. in Computer Science from T el Aviv University, after which he spent three years at the Computer Scien ce Robotics Laboratory at Stanford University. He then joined the Departme nt of Computer Science at Tel Aviv University, where he is currently a ful l professor and for two years was the department chair. Halperin’s main fi eld of research is Computational Geometry and Its Applications. Applicatio n areas he is interested in include robotics, automated manufacturing, alg orithmic motion planning, and 3D printing. A major focus of Halperin’s wor k has been in research and development of robust geometric software, in co llaboration 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 ro botics, including SoCG, WAFR, ESA, and ALENEX. Halperin is an ACM Fellow a nd an IEEE Fellow. ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il LOCATION:Room 337 taub bld. UID:123se24012024104010 END:VEVENT BEGIN:VEVENT DTSTART;TZID="Asia/Jerusalem":20230509T180000 DTEND;TZID="Asia/Jerusalem":20230509T200000 DTSTAMP;TZID="Asia/Jerusalem":20230509T180000 FREEBUSY;FBTYPE=BUSY:20230509T180000/20230509T200000 SUMMARY;LANGUAGE=en-US:CSpecial Event about StarkWare Tech Talk at 2023-05 -09 18:00:00 DESCRIPTION;LANGUAGE=en-US: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. Please pre-register. ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il LOCATION:Junta Bar, Technion UID:123se24012024104050 END:VEVENT BEGIN:VEVENT DTSTART;TZID="Asia/Jerusalem":20230510T123000 DTEND;TZID="Asia/Jerusalem":20230510T143000 DTSTAMP;TZID="Asia/Jerusalem":20230510T123000 FREEBUSY;FBTYPE=BUSY:20230510T123000/20230510T143000 SUMMARY;LANGUAGE=en-US:CSpecial Event about Recruitment Day by Intuit at 2 023-05-10 12:30:00 DESCRIPTION;LANGUAGE=en-US:Intuit - a global fintech company - will hold a recruitment day and will present its business in Trust Data & Deep Insigh t, 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:123se24012024104080 END:VEVENT BEGIN:VEVENT DTSTART;TZID="Asia/Jerusalem":20230510T123000 DTEND;TZID="Asia/Jerusalem":20230510T133000 DTSTAMP;TZID="Asia/Jerusalem":20230510T123000 FREEBUSY;FBTYPE=BUSY:20230510T123000/20230510T133000 SUMMARY;LANGUAGE=en-US:Theory Seminar talk by Amnon Ta-Shma (Tel-Aviv univ ersity) about Theory Seminar: HDX Condensers at 2023-05-10 12:30:00 DESCRIPTION;LANGUAGE=en-US:More than twenty years ago, Capalbo, Reingold, Vadhan and Wigderson gave the first (and up to date only) explicit constru ction of a bipartite expander with almost full combinatorial expansion. Th e construction incorporates zig-zag ideas together with extractor technolo gy, and is rather complicated. We give an alternative construction that bu ilds upon recent constructions of hyper-regular, high-dimensional expander s. The new construction is, in our opinion, simple and elegant. Beyond demonstrating a new, surprising, and intriguing, application of high-dimen sional expanders, the construction employs new ideas which we hope may lea d to progress on the still remaining open problems in the area Joint work with Itay Cohen and Roy Roth from Tel-Aviv University. ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il LOCATION:Taub 201 UID:123se24012024104120 END:VEVENT BEGIN:VEVENT DTSTART;TZID="Asia/Jerusalem":20230510T183000 DTEND;TZID="Asia/Jerusalem":20230510T203000 DTSTAMP;TZID="Asia/Jerusalem":20230510T183000 FREEBUSY;FBTYPE=BUSY:20230510T183000/20230510T203000 SUMMARY;LANGUAGE=en-US:CSpecial Event about Practical Information Gatherin g Workshop by CYE at 2023-05-10 18:30:00 DESCRIPTION;LANGUAGE=en-US:You are invited to an information gathering wor kshop 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 pro cess of identifying vulnerabilities in the organization, about tools and t echniques for identifying organizational assets exposed to the Internet, a nd about gathering information using familiar tools and their use of the l eading technologies in the market, on Wednesday, May 10, 2023, 18:30 at Ta ub 337. More
details and pre-registration - the number of places is li mited, participation is subject to confirmation of registration. ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il LOCATION:Room 337 taub bld. UID:123se24012024104100 END:VEVENT BEGIN:VEVENT DTSTART;TZID="Asia/Jerusalem":20230514T113000 DTEND;TZID="Asia/Jerusalem":20230514T123000 DTSTAMP;TZID="Asia/Jerusalem":20230514T113000 FREEBUSY;FBTYPE=BUSY:20230514T113000/20230514T123000 SUMMARY;LANGUAGE=en-US:TDC talk by Armando Castañeda (National Autonomous University of Mexico) about Distributed Computing Seminar: Asynchronous Wa it-Free Runtime Verification and Enforcement of Linearizability at 2023-05 -14 11:30:00 DESCRIPTION;LANGUAGE=en-US:This work studies the problem of distributed ru ntime verification of linearizability for asynchronous concurrent implemen tations. It proposes an interactive model for distributed runtime verifica tion 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 c aptures 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:123se24012024104180 END:VEVENT BEGIN:VEVENT DTSTART;TZID="Asia/Jerusalem":20230514T143000 DTEND;TZID="Asia/Jerusalem":20230514T153000 DTSTAMP;TZID="Asia/Jerusalem":20230514T143000 FREEBUSY;FBTYPE=BUSY:20230514T143000/20230514T153000 SUMMARY;LANGUAGE=en-US:Coding_Theory_Semina talk by Daniella Bar-Lev (CS, Technion) about Coding Theory: Cover Your Bases: How to Minimize the Seque ncing Coverage in DNA Storage Systems at 2023-05-14 14:30:00 DESCRIPTION;LANGUAGE=en-US:This seminar will be divided into two parts. In the first part, we will provide an introduction to DNA storage systems. T his will include an overview of their biological and computational compone nts, as well as a survey of the current technologies and emerging trends i n the market landscape. In the second part, we will focus on a novel p roblem called the DNA coverage depth problem. Motivated by the high cost a nd 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 dep th as a function of the DNA storage channel, the error-correcting code, an d the reconstruction algorithm. We will study the DNA coverage depth probl em under both random and non-random access settings and explore coding sch emes that optimize the required coverage depth. Daniella Bar-Lev is a P h.D. student in the Computer Science Department at the Technion -- Israel Institute of Technology. She is a recipient of the Gutwirth Excellence Sch olarship 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, respectiv ely. Her research interests include algorithms, discrete mathematics, codi ng theory, and DNA storage ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il LOCATION:Taub 601 UID:123se24012024104150 END:VEVENT BEGIN:VEVENT DTSTART;TZID="Asia/Jerusalem":20230516T113000 DTEND;TZID="Asia/Jerusalem":20230516T123000 DTSTAMP;TZID="Asia/Jerusalem":20230516T113000 FREEBUSY;FBTYPE=BUSY:20230516T113000/20230516T123000 SUMMARY;LANGUAGE=en-US:pixel-club talk by Niv Cohen (Hebrew University of Jerusalem) about Pixel Club: The Success and Challenges of Representation- Based Anomaly Detection at 2023-05-16 11:30:00 DESCRIPTION;LANGUAGE=en-US:Anomaly detection aims to discover data which d iffer from the norm in a semantically meaningful manner. The task is diffi cult as anomalies are rare and unexpected. Moreover, a sample can be an im portant anomaly to one person and an uninteresting statistical outlier to another. In this talk, I will first present how deep representations br ought substantial gains for image anomaly detection and segmentation. Next , we will discuss the types of representations that are beneficial for ano maly detection, and how a given representation can be improved. Finally, I will present two remaining challenges and initial directions toward addre ssing 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. Bio: Niv is a Ph.D. student at the Hebr ew University of Jerusalem, advised by Dr. Yedid Hoshen. He received his B Sc. in mathematics with physics, and M.Sc. in physics, both from the Techn ion. He's interested in computer vision and representation learning with a focus on anomaly detection and scientific data. ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il LOCATION:Room 1061, EE Meyer Building UID:123se24012024104160 END:VEVENT BEGIN:VEVENT DTSTART;TZID="Asia/Jerusalem":20230517T123000 DTEND;TZID="Asia/Jerusalem":20230517T133000 DTSTAMP;TZID="Asia/Jerusalem":20230517T123000 FREEBUSY;FBTYPE=BUSY:20230517T123000/20230517T133000 SUMMARY;LANGUAGE=en-US:colloq talk by Thomas Vidick (Weizmann Institute of Science) about CS Colloquia: Testing Quantum Systems in the High-complexi ty Regime at 2023-05-17 12:30:00 DESCRIPTION;LANGUAGE=en-US:From carefully crafted quantum algorithms to in formation-theoretic security in cryptography, a quantum computer can achie ve impressive feats with no classical analogue. Can their correct realizat ion be verified? When the power of the device greatly surpasses that of th e user, computationally as well as cryptographically, what means of contro l remain available to the user? Recent lines of work in quantum cryptograp hy and complexity develop approaches to this question based on the notion of an interactive proof. Generally speaking an interactive proof models an y interaction whereby a powerful device aims to convince a restricted user of the validity of an agree-upon statement -- such as that the machine ge nerates perfect random numbers or executes a specific quantum algorithm. T wo models have emerged in which large-scale verification has been shown po ssible: either by placing reasonable computational assumptions on the quan tum device, or by requiring that it consists of isolated components across which Bell tests can be performed. In the talk I will discuss recent a dvances on the verification power of interactive proof systems between a q uantum device and a classical user, focusing on the certification of quant um randomness from a single device, under cryptographic assumptions. B io: Thomas Vidick is professor of Computer Science at the Weizmann Inst itute of Science, which he joined in 2022. Between 2014 and 2022 he was As sistant Professor, and then Professor, at the California Institute of Tech nology. Prior to joining Caltech, Vidick earned a B.A. in pure mathematics from Ecole Normale Superieure in Paris, a Masters in Computer Science fro m Universite Paris 7 and a Ph.D. from UC Berkeley. In 2020-2022 he was a postdoctoral scholar at the Massachusetts Institute of Technology, supervi sed by Scott Aaronson. Vidick's Ph.D. thesis was awarded the Bernard Fr iedman memorial prize in applied mathematics. In 2017 he was named a CIFAR Azrieli Global Scholar. In 2019 he received a Presidential Early-Career A ward (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. V idick's research is situated at the interface of theoretical computer scie nce, quantum information and cryptography. He has investigated the role of entanglement in multiprover interactive proof systems and in quantum cryp tography, making important contributions to both areas. ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il LOCATION:Room 337 taub bld. UID:123se24012024104020 END:VEVENT BEGIN:VEVENT DTSTART;TZID="Asia/Jerusalem":20230517T123000 DTEND;TZID="Asia/Jerusalem":20230517T143000 DTSTAMP;TZID="Asia/Jerusalem":20230517T123000 FREEBUSY;FBTYPE=BUSY:20230517T123000/20230517T143000 SUMMARY;LANGUAGE=en-US:CSpecial Event about Recruitment Day by Istra Resea rch at 2023-05-17 12:30:00 DESCRIPTION;LANGUAGE=en-US:Istra Research will hold a r ecruitment 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 "Introduction to Algorithm Trading" - High Frequency Trading - that you review basic concepts i n the field. Please register in advance Registration for the Istra R iddle (announcement of the winner of the riddle and the prize will tak e place on May 31, 2023). ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il LOCATION:CS Taub Lobby and Taub 3 UID:123se24012024104130 END:VEVENT BEGIN:VEVENT DTSTART;TZID="Asia/Jerusalem":20230517T173000 DTEND;TZID="Asia/Jerusalem":20230517T193000 DTSTAMP;TZID="Asia/Jerusalem":20230517T173000 FREEBUSY;FBTYPE=BUSY:20230517T173000/20230517T193000 SUMMARY;LANGUAGE=en-US:CSpecial Event about Round Tabels Event by Intel at 2023-05-17 17:30:00 DESCRIPTION;LANGUAGE=en-US:You are invited to a round tables event with In tel 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. Please pre-register. ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il LOCATION:Room 337 taub bld. UID:123se24012024104140 END:VEVENT BEGIN:VEVENT DTSTART;TZID="Asia/Jerusalem":20230522T130000 DTEND;TZID="Asia/Jerusalem":20230522T140000 DTSTAMP;TZID="Asia/Jerusalem":20230522T130000 FREEBUSY;FBTYPE=BUSY:20230522T130000/20230522T140000 SUMMARY;LANGUAGE=en-US:phd talk by Hadar Sivan about Efficient First and S econd Order Methods for Function Monitoring and Optimization at 2023-05-22 13:00:00 DESCRIPTION;LANGUAGE=en-US:Machine learning model training is a computatio nally expensive task that requires significant amounts of time and resourc es, especially for larger models. The problem is further increased when da ta arrives in a continuous stream since the model must be retrained multip le times to incorporate the new data and ensure the model remains accurate . Another difficulty arises during inference time when the data is geo-dis tributed; 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 o f the function to reduce optimization time and network overhead. Our propo sed solutions aim to improve the efficiency and scalability of machine lea rning 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:123se24012024104070 END:VEVENT BEGIN:VEVENT DTSTART;TZID="Asia/Jerusalem":20230523T113000 DTEND;TZID="Asia/Jerusalem":20230523T123000 DTSTAMP;TZID="Asia/Jerusalem":20230523T113000 FREEBUSY;FBTYPE=BUSY:20230523T113000/20230523T123000 SUMMARY;LANGUAGE=en-US:pixel-club talk by Gilad Lerman (University of Minn esota) about Pixel Club: Cycle-edge Message Passing for Group and Non-grou p Synchronization at 2023-05-23 11:30:00 DESCRIPTION;LANGUAGE=en-US:The general synchronization problem asks to rec over states of objects from their corrupted relative measurements. When th e states are represented by group elements (e.g. 3-D rotations or permutat ions) this problem is known as group synchronization. In several applicati ons, the algebraic structure of the states is more complicated, for exampl e, the states can be represented by partial permutations. The synchronizat ion 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 ex plain its generalization to non-groups, by exemplifying the case of partia l permutation synchronization. I will emphasize mathematical difficulties, review some mathematical guarantees for the proposed methods and also dem onstrate 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:123se24012024104220 END:VEVENT BEGIN:VEVENT DTSTART;TZID="Asia/Jerusalem":20230524T113000 DTEND;TZID="Asia/Jerusalem":20230524T123000 DTSTAMP;TZID="Asia/Jerusalem":20230524T113000 FREEBUSY;FBTYPE=BUSY:20230524T113000/20230524T123000 SUMMARY;LANGUAGE=en-US:ceClub talk by Miron Livny (University of Wisconsin -Madison) about ceClub: Translational Computer Science at Work at 2023-05- 24 11:30:00 DESCRIPTION;LANGUAGE=en-US:The UW-Madison Center for High Throughput Compu ting (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 togethe r innovation in distributed computing and services to scientists. Evaluati on of new technologies under real-life conditions by engaged users advance d scientific discovery and guided the center in future research and develo pment activities. The HTCSS has been serving for almost four decades as a means to place new capabilities in the hands of researchers who can benefi t from advances in Throughput Computing. Having a capable and dependable s uite of software tools enabled adoption of advanced Throughput Computing m ethodologies by an international community of users. Abramson and Parashar in their recent formalization of Translational Research in Computer Scien ce (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 technol ogy is deployed and evaluated. The talk will present the different elem ents that facilitate the translational work of the CHTC. These include a c ampus wide and a national research computing environment and a sustained s equence of software releases. The challenges of sustaining such a center i n an academic institution will be discussed and the opportunities for inno vation triggered by ever evolving research computing environments at all s cales will be reviewed. Bio: Miron Livny received a B.Sc. degree in Ph ysics 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 1 978 and 1984, respectively. Since 1983 he has been on the Computer Science s Department faculty at the University of Wisconsin-Madison, where he is c urrently the John P. Morgridge Professor of Computer Science. He serves as the director of the Center for High Throughput Computing (CHTC), is leadi ng 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 Morgr idge Institute of Research where he leads the Research Computing theme. Dr. Livny’s research focuses on distributed processing and data managemen t systems and involves close collaboration with researchers from a wide sp ectrum of disciplines. He pioneered the area of High Throughput Computing (HTC) and developed frameworks and software tools that have been widely ad opted by academic and commercial organizations around the world. Livny is the recipient of the 2006 ACM SIGMOD Test of Time Award the 2013 HPDC A chievement Award and the 2020 IEEE TCDP Outstanding Technical Achievement Award. ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il LOCATION:Taub 401 UID:123se24012024104230 END:VEVENT BEGIN:VEVENT DTSTART;TZID="Asia/Jerusalem":20230524T123000 DTEND;TZID="Asia/Jerusalem":20230524T133000 DTSTAMP;TZID="Asia/Jerusalem":20230524T123000 FREEBUSY;FBTYPE=BUSY:20230524T123000/20230524T133000 SUMMARY;LANGUAGE=en-US:Theory Seminar talk by Uriya First (Haifa universit y) about Theory Seminar: A Sheaf-theoretic Approach to Constructing Locall y Testable Codes at 2023-05-24 12:30:00 DESCRIPTION;LANGUAGE=en-US:I will discuss a new approach towards construct ing good locally testable codes (LTCs) with better qualities than the rece nt 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 conditi ons and a cohomological condition --- both of which can be checked in fini te (constant) time ---, then this example could be propagated into an infi nite family of good LTCs. We also propose a heuristic method for construct ing the initial sheaved HDX. The LTCs arising from our framework are 2-que ry LTCs which also admit a stronger testability property called (T)-testab ility. This is a joint work with Tali Kaufman. ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il LOCATION:Taub 201 UID:123se24012024104280 END:VEVENT BEGIN:VEVENT DTSTART;TZID="Asia/Jerusalem":20230524T130000 DTEND;TZID="Asia/Jerusalem":20230524T170000 DTSTAMP;TZID="Asia/Jerusalem":20230524T130000 FREEBUSY;FBTYPE=BUSY:20230524T130000/20230524T170000 SUMMARY;LANGUAGE=en-US:CSpecial Event about Excellence Program Alumni Conf erence at 2023-05-24 13:00:00 DESCRIPTION;LANGUAGE=en-US:You are invited to celebrate 30 years of excell ence: the Technion program for excellence is celebrating 30 years since it s establishment in a series of lectures, on Wednesday, May 24, 2023 betwee n 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 o n the trends, The latest innovations and challenges in their field. Amo ng 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. After the lectures, there will be a discussion on the subject of academia, industry and the defense system with the participation of Prof. Hagit Mes ser-Yeron, Aharon Aharon and Eyal Hulata, and moderator Chen Lieberman. Details and ful l program. The 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:123se24012024104250 END:VEVENT BEGIN:VEVENT DTSTART;TZID="Asia/Jerusalem":20230528T120000 DTEND;TZID="Asia/Jerusalem":20230528T130000 DTSTAMP;TZID="Asia/Jerusalem":20230528T120000 FREEBUSY;FBTYPE=BUSY:20230528T120000/20230528T130000 SUMMARY;LANGUAGE=en-US:msc talk by Natan Kaminsky about Lead Optimization For Drug Discovery With Limited Data at 2023-05-28 12:00:00 DESCRIPTION;LANGUAGE=en-US:Drug development is a long and costly process c onsisting of several stages that can take many years to complete. One of the early stage's goals is to optimize a novel chemical compound to be active against a target protein associated with the disease. The goal of molecule optimization is, given an input molecule, to produce a new mol ecule that is chemically similar to the input molecule but with an improve d property. In this work, we present a novel approach for optimizing m olecules. We propose to represent a molecule by breaking it into two disjo int substructures that we call: the molecule chains and the molecule core. We train a model to generate the molecule chains with the desired pro perties for optimization, which are then attached to the molecule core to construct a novel molecule with high similarity to the input molecule. We then show how to extend this approach to tasks where data is scarce, su ch 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:123se24012024104190 END:VEVENT BEGIN:VEVENT DTSTART;TZID="Asia/Jerusalem":20230528T143000 DTEND;TZID="Asia/Jerusalem":20230528T153000 DTSTAMP;TZID="Asia/Jerusalem":20230528T143000 FREEBUSY;FBTYPE=BUSY:20230528T143000/20230528T153000 SUMMARY;LANGUAGE=en-US:Coding_Theory_Semina talk by Christoph Hofmeister ( Technical University of Munich (TUM), Munich, Germany) about Coding Theory : How to Catch Liars in distributed Gradient Descent at 2023-05-28 14:30:0 0 DESCRIPTION;LANGUAGE=en-US:This talk is about distributed machine learning in the presence of Byzantine errors. A main node performs gradient descen t steps with the help of some worker nodes, a limited number of which are controlled by an adversary. These malicious worker nodes can return arbitr ary data to the main node instead of the desired computation results. Prio r work proposes distributing the data with redundancy among the workers an d using error correction codes to detect and correct the erroneous computa tion results. In this work, we propose a solution that requires less redun dancy at the cost of a small number of gradient computations on the main n ode and some light communication. In addition to a new scheme, we provide lower bounds on communication and computation. Christoph Hofmeister rec eived a B.Eng. in electrical engineering and information technology from t he Munich University of Applied Sciences (HM) in 2019 and an M.Sc. in elec trical engineering and information technology from the Technical Universit y of Munich (TUM) in 2021, where he is currently pursuing a Ph.D. with the Coding and Cryptography Group, Institute of Communications Engineering, u nder the supervision of Prof. Wachter-Zeh. His research interests include information and coding theory and its applications, with a focus on coded computing. ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il LOCATION:Taub 601 UID:123se24012024104290 END:VEVENT BEGIN:VEVENT DTSTART;TZID="Asia/Jerusalem":20230530T090000 DTEND;TZID="Asia/Jerusalem":20230530T150000 DTSTAMP;TZID="Asia/Jerusalem":20230530T090000 FREEBUSY;FBTYPE=BUSY:20230530T090000/20230530T150000 SUMMARY;LANGUAGE=en-US:CSpecial Event about Intel's Tech Experience at 202 3-05-30 09:00:00 DESCRIPTION;LANGUAGE=en-US:You are invited to Intel's Tech Experience event, on Tuesday, May 30, 2023, on the "Shany" Plaza of Taub Bu ilding: Between 9:00-12:00 - AR/VR complex Between 12:00-13:30 / 13:30-1 5:00 - two rounds of the FPGA workshop Hello world More details, full program and pre-registrati on ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il LOCATION:"Shany" Plaza of Taub Building UID:123se24012024104380 END:VEVENT BEGIN:VEVENT DTSTART;TZID="Asia/Jerusalem":20230531T123000 DTEND;TZID="Asia/Jerusalem":20230531T143000 DTSTAMP;TZID="Asia/Jerusalem":20230531T123000 FREEBUSY;FBTYPE=BUSY:20230531T123000/20230531T143000 SUMMARY;LANGUAGE=en-US:CSpecial Event about Recruitment Day by CYE at 2023 -05-31 12:30:00 DESCRIPTION;LANGUAGE=en-US:You are invited to recruitment day by CYE with engineers and recruitment teams and to a technological lecture by Dr. Nimr od 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 org anizations (and what do you do when the problem is NP-hard)? - on Wednesda y, May 31, 2023 at 12:30 in the Taub lobby, and the lecture will take plac e at 13:30 in the Taub 012 Auditorium in the Learning Center on the entran ce floor. Please pre-register. ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il LOCATION:Taub 012 (Learning Center Auditorium) UID:123se24012024104330 END:VEVENT BEGIN:VEVENT DTSTART;TZID="Asia/Jerusalem":20230531T123000 DTEND;TZID="Asia/Jerusalem":20230531T133000 DTSTAMP;TZID="Asia/Jerusalem":20230531T123000 FREEBUSY;FBTYPE=BUSY:20230531T123000/20230531T133000 SUMMARY;LANGUAGE=en-US:Theory Seminar talk by Dean Doron (Ben-Gurion Unive rsity) about Theory Seminar: Almost Chor–Goldreich Sources and Adversarial Random Walks at 2023-05-31 12:30:00 DESCRIPTION;LANGUAGE=en-US:In this talk we consider the following adversar ial, non-Markovian, random walk on “good enough” expanders: Starting from some fixed vertex, walk according to the instructions X = X1,…,Xt, where e ach Xi is only somewhat close to having only little entropy, conditioned o n 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. We show that such walks (or certain vari ants of them) accumulate most of the entropy in X. We call such X-s “al most Chor–Goldreich (CG) Sources”, and our result gives deterministic cond ensers 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 Sant ha–Vazirani sources. As a consequence, we can simulate any randomized algorithm with small failure probability using almost CG sources with no m ultiplicative slowdown. This result extends to randomized protocols as wel l, and any setting in which we cannot simply cycle over all seeds, and a “ one-shot” simulation is needed. Joint work with Dana Moshkovitz, Justin Oh, and David Zuckerman ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il LOCATION:Taub 201 UID:123se24012024104360 END:VEVENT BEGIN:VEVENT DTSTART;TZID="Asia/Jerusalem":20230531T143000 DTEND;TZID="Asia/Jerusalem":20230531T153000 DTSTAMP;TZID="Asia/Jerusalem":20230531T143000 FREEBUSY;FBTYPE=BUSY:20230531T143000/20230531T153000 SUMMARY;LANGUAGE=en-US:phd talk by Idan Yaniv about Improving the Performa nce and Evaluation Methodology of Virtual Memory Systems at 2023-05-31 14: 30:00 DESCRIPTION;LANGUAGE=en-US:The virtual memory subsystem translates the add ress of each memory reference from its virtual to its physical representat ion, increasing execution runtimes by as much as 50% and 90% in bare-metal and virtual setups, respectively. We alleviate these overheads by develop ing improved virtual memory designs: (i) hashed page tables and (ii) TLB p artitioning for simultaneous multithreading. We additionally develop an ef ficient and reliable methodology for evaluating the performance of newly p roposed 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 pr edicts the execution runtime. ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il LOCATION:Zoom Lecture and Taub 601 UID:123se24012024104240 END:VEVENT BEGIN:VEVENT DTSTART;TZID="Asia/Jerusalem":20230531T163000 DTEND;TZID="Asia/Jerusalem":20230531T173000 DTSTAMP;TZID="Asia/Jerusalem":20230531T163000 FREEBUSY;FBTYPE=BUSY:20230531T163000/20230531T173000 SUMMARY;LANGUAGE=en-US:msc talk by Boaz Moav about Tail-Erasure-Correcting Codes at 2023-05-31 16:30:00 DESCRIPTION;LANGUAGE=en-US:The increasing demand for data storage has prom pted the exploration of new techniques, with molecular data storage being a promising alternative. The stored information can be represented as a co llection of two-dimensional arrays, such that each row represents a DNA st rand. In this work, we present the results of our research into error-corr ecting codes for molecular data storage using this representation. Althoug h 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 occu rs when a defective bit prematurely terminates the chain, or the data with in a cell can be corrupted completely. Initial experiments have reported e rror rates as high as 10%. Those errors can be considered as erasures in t he last few symbols. Our study focuses on correcting those erasures and al so deletions across rows. We present code constructions and explicit encod ers 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 ch osen parity check matrix, such that the set of linearly independent column s match the pattern of possible erasures, that is at most a known paramete r $d$ of tail-erasures. This construction is optimal for $d=2,3,4$ and nea rly 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 d istance, but which is better suited for the tail-erasure model. The second construction uses Tensor Product codes (TPC), that underlies Varshamov-Te nengolts (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 pr oblems, to construct a code that can correct both tail-erasures and one de letion 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:123se24012024104110 END:VEVENT BEGIN:VEVENT DTSTART;TZID="Asia/Jerusalem":20230531T190000 DTEND;TZID="Asia/Jerusalem":20230531T210000 DTSTAMP;TZID="Asia/Jerusalem":20230531T190000 FREEBUSY;FBTYPE=BUSY:20230531T190000/20230531T210000 SUMMARY;LANGUAGE=en-US:CSpecial Event about "Research on the Bar" Evening - TED Lectures at 2023-05-31 19:00:00 DESCRIPTION;LANGUAGE=en-US:You are invited to the "Research on the Bar" ev ening - TED lectures and a meeting with three faculty members and their re search groups on Wednesday, May 31, 2023 at 19:00 pm in Taub Terrace: P rof. Eitan Yacobi: Storing information in DNA: Who ate my files? Dr. Shau l Almagor: How to never make a mistake in anything Dr. Ron Rothblum: How to prove without revealing anything Please pre-register. ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il LOCATION:Taub Terrace UID:123se24012024104270 END:VEVENT BEGIN:VEVENT DTSTART;TZID="Asia/Jerusalem":20230601T110000 DTEND;TZID="Asia/Jerusalem":20230601T120000 DTSTAMP;TZID="Asia/Jerusalem":20230601T110000 FREEBUSY;FBTYPE=BUSY:20230601T110000/20230601T120000 SUMMARY;LANGUAGE=en-US:msc talk by Yara Shamshoum about Measuring The Com plexity of Neural Network Algorithms at 2023-06-01 11:00:00 DESCRIPTION;LANGUAGE=en-US:Substantial efforts have been devoted into impr oving the capabilities of neural networks to solve algorithmic tasks. Thro ugh training, these networks learn to mimic algorithmic behaviour, enablin g them to handle tasks such as sorting, navigating, and managing complex d ata structures like graphs. However, classic algorithms and neural network s are fundamentally different, making it challenging to analyze the comple xity of an algorithm learned by a neural network. First, it is necessary t o establish that the model has indeed learned an abstract solution resembl ing algorithmic behaviour. Additionally, while the complexity of classic a lgorithms is measured asymptotically, the complexity of an algorithm learn ed by a neural network must be measured within the finite input space supp orted 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 meth ods 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:123se24012024104390 END:VEVENT BEGIN:VEVENT DTSTART;TZID="Asia/Jerusalem":20230605T084500 DTEND;TZID="Asia/Jerusalem":20230607T210000 DTSTAMP;TZID="Asia/Jerusalem":20230605T084500 FREEBUSY;FBTYPE=BUSY:20230605T084500/20230607T210000 SUMMARY;LANGUAGE=en-US:CSpecial Event about SYSTOR 2023 at 2023-06-05 08:4 5:00 DESCRIPTION;LANGUAGE=en-US:You are invited to participate in the internati onal conference SYSTOR 2023, leader in the fields of systems, cloud and st orage, 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. Participation is free of charge but requires pre-registration. More details. ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il LOCATION:Technion Samuel Neaman Institute UID:123se24012024104200 END:VEVENT BEGIN:VEVENT DTSTART;TZID="Asia/Jerusalem":20230605T183000 DTEND;TZID="Asia/Jerusalem":20230605T203000 DTSTAMP;TZID="Asia/Jerusalem":20230605T183000 FREEBUSY;FBTYPE=BUSY:20230605T183000/20230605T203000 SUMMARY;LANGUAGE=en-US:CSpecial Event about NLP Research Session at 2023-0 6-05 18:30:00 DESCRIPTION;LANGUAGE=en-US:You are invited to the research meeting - NLP R esearch Night (in collaboration with Grove Ventures - round tables with senior research ers from academia and industry in an open dialogue about the topics, chall enges and burning questions in th e field of NLP, on Monday, June 5, 2023, 18:30 in Taub Terrace. Pre-registration is requi red, and is subject to approval. ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il LOCATION:Taub Terrace UID:123se24012024104320 END:VEVENT BEGIN:VEVENT DTSTART;TZID="Asia/Jerusalem":20230607T113000 DTEND;TZID="Asia/Jerusalem":20230607T123000 DTSTAMP;TZID="Asia/Jerusalem":20230607T113000 FREEBUSY;FBTYPE=BUSY:20230607T113000/20230607T123000 SUMMARY;LANGUAGE=en-US:ceClub talk by Sam Noh (Virginia Tech) about ceClub : DyTIS: A Dynamic Dataset Targeted Index Structure Simultaneously Efficie nt for Search, Insert, and Scan at 2023-06-07 11:30:00 DESCRIPTION;LANGUAGE=en-US:Many datasets in real life are complex and dyna mic, 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 s tructure 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 data set, and learns and adjusts its structure as the dataset grows. The key no velty 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 ha sh 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 lea rned index for the dynamic datasets considered. Bio: Sam H.(Hyuk) Noh received the BE degree in computer engineering from the Seoul National Uni versity, Seoul, Korea, in 1986, and the PhD degree from the Department of Computer Science, University of Maryland, College Park, MD, in 1993. He he ld a visiting faculty position at the George Washington University, Washin gton, DC, from 1993 to 1994 before joining Hongik University, in Seoul, Ko rea, where he was a professor in the School of Computer and Information En gineering until the Spring of 2015. During this period, he served as the C hair 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 t o August 2002, he was also a visiting associate professor with the Univers ity of Maryland Institute of Advanced Computer Studies (UMIACS), College P ark, MD. Starting from the Fall of 2015 he joined UNIST (Ulsan National In stitute of Science and Technology), a young science and tech focused natio nal university, where he was a Professor at the Department of Computer Sci ence and Engineering and served as the inaugural Dean of the Graduate Scho ol of Artificial Intelligence in the College of Information and Biotechnol ogy from August 2020 through March 2023. He also served as the Dean of the School of Electrical and Computer Engineering from January of 2016 throug h June of 2018. As of January 2023, he is a Professor at the Computer Scie nce Department at Virginia Tech. He has served/serves as General Chair, Pr ogram Chair, and Program Committee Member on a number of technical confere nces and workshops. He also served on the Steering Committee of LCTES from 2016 through 2020. He is currently the Chair of the Steering Committee fo r ACM HotStorage and a Steering Committee member of USENIX FAST and IEEE N VMSA. He also served as Editor-in-Chief of the ACM Transactions on Storage from 2016 through 2022. His research interests include system software is sues pertaining to computer systems in general and storage systems in part icular, with a focus on the use of new memory technologies such as flash m emory and persistent memory. He is a Fellow of the ACM and IEEE and a memb er of USENIX and KIISE (Korean Institute of Information Scientists and Eng ineers). ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il LOCATION:Taub 401 UID:123se24012024104410 END:VEVENT BEGIN:VEVENT DTSTART;TZID="Asia/Jerusalem":20230607T123000 DTEND;TZID="Asia/Jerusalem":20230607T133000 DTSTAMP;TZID="Asia/Jerusalem":20230607T123000 FREEBUSY;FBTYPE=BUSY:20230607T123000/20230607T133000 SUMMARY;LANGUAGE=en-US:msc talk by Adi Amuzig about Value of Assistance fo r Mobile Agents at 2023-06-07 12:30:00 DESCRIPTION;LANGUAGE=en-US:Mobile robotic agents often suffer from localiz ation uncertainty which grows with time and with the agents' movement. Thi s 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 b e 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 a nd 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 co st 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 u ncertainty, modeled as a Gaussian process. We specify conditions under whi ch our VOA measure is valid, and empirically demonstrate the ability of ou r measure to predict the agent's average cost reduction when receiving ass istance in both simulated and real-world robotic settings. ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il LOCATION:Zoom Lecture: 92471871959 UID:123se24012024104210 END:VEVENT BEGIN:VEVENT DTSTART;TZID="Asia/Jerusalem":20230607T123000 DTEND;TZID="Asia/Jerusalem":20230607T133000 DTSTAMP;TZID="Asia/Jerusalem":20230607T123000 FREEBUSY;FBTYPE=BUSY:20230607T123000/20230607T133000 SUMMARY;LANGUAGE=en-US:Theory Seminar talk by Roy Gotlib (Bar-Ilan Univers ity) about Theory Seminar: List Agreement Testing and High Dimensional Exp ansion at 2023-06-07 12:30:00 DESCRIPTION;LANGUAGE=en-US:One of the key components in PCP constructions are agreement tests. In agreement testing the tester is given access to subsets of fixed size of some set, each equipped with an assignment. T he tester is then tasked with testing whether these local assignments agre e with some global assignment over the entire set. One natural generali zation of this concept is the case where, instead of a single assignment t o each local view, the tester is given access to $\ell$ different assignme nts for every subset. The tester is then tasked with testing whether th ere exist $\ell$ global functions that agree with all of the assignments o f all of the local views. In this talk I will present a recent result t hat shows that if the subsets form a (sufficiently good) high dimensional expander then they support list agreement testing under mild assumptions o n the local assignments. In addition, I will also show that those mild assumptions are necessary for list agreement. I will not assume any pri or knowledge of high dimensional expanders. Based on a joint work with Tali Kaufman ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il LOCATION:Taub 201 UID:123se24012024104430 END:VEVENT BEGIN:VEVENT DTSTART;TZID="Asia/Jerusalem":20230607T153000 DTEND;TZID="Asia/Jerusalem":20230607T163000 DTSTAMP;TZID="Asia/Jerusalem":20230607T153000 FREEBUSY;FBTYPE=BUSY:20230607T153000/20230607T163000 SUMMARY;LANGUAGE=en-US:ceClub talk by Yuli Mandelblat (Intel) about ceClub : Recent Intel Hybrid CPU Architecture at 2023-06-07 15:30:00 DESCRIPTION;LANGUAGE=en-US:Why hybrid and what problems this technology so lves? What is the difference between Intel’s Hybrid Technology and the ot her market solutions (e.g. Big-little)? Hybrid micro-architectural soluti ons: caches, fabric. How SW knows what core to use for what task. Intel Thread Director - what is does and why it is required. Future development of Hybrid solutions. Bio: Yuli (Julius) Mandelblat is an Intel Fellow of Client SoC Architecture Team. Yuli works at Intel since 1990. Throug hout his career, he worked on multiple Intel CPUs in variety of areas of p roduct development from validation and design to architecture. He was resp onsible for the development of the first multi-core solutions, on-die inte rconnect, memory ordering and coherency solutions, etc. In the latest Inte l® client products Yuli led the definition of highly successful Hybrid arc hitecture that was a key for the success of CPU’s of 12th and 13th generat ions of Intel CPU. Yuli holds M.Sc. degree from Russian University of Tra nsport (MIIT). Yuli is an inventor of more than 20 patents in different a reas of CPU architecture. ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il LOCATION:Room 352, EE Meyer Building UID:123se24012024104420 END:VEVENT BEGIN:VEVENT DTSTART;TZID="Asia/Jerusalem":20230607T163000 DTEND;TZID="Asia/Jerusalem":20230607T173000 DTSTAMP;TZID="Asia/Jerusalem":20230607T163000 FREEBUSY;FBTYPE=BUSY:20230607T163000/20230607T173000 SUMMARY;LANGUAGE=en-US:phd talk by Maria Abu Sini about The Reconstruction Model and Shortmers-based DNA Synthesis at 2023-06-07 16:30:00 DESCRIPTION;LANGUAGE=en-US:Levenshtein's reconstruction model was first in troduced in 2001 and suggests transmitting a word over multiple noisy chan nels, then using the channels' outputs to recover the transmitted word. Th is talk will discuss the reconstruction model when the channels are prone to combinations of errors, or when unique retrieval of the transmitted wor d 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 shor tmers (i.e., sequences of bases) will be investigated from a theoretical p oint of view. This optimization differs from the conventional synthesis pr ocess by appending in each cycle not only single bases, but also shortmers . The significance of this optimization lies in reducing the number of cyc les, which then determines the time and monetary cost of the synthesis pro cess. Hence, the talk will discuss several questions pertaining to this op timization, such as which shortmers to use and how to calculate the minimu m number of cycles. ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il LOCATION:Taub 601 UID:123se24012024104300 END:VEVENT BEGIN:VEVENT DTSTART;TZID="Asia/Jerusalem":20230608T110000 DTEND;TZID="Asia/Jerusalem":20230608T120000 DTSTAMP;TZID="Asia/Jerusalem":20230608T110000 FREEBUSY;FBTYPE=BUSY:20230608T110000/20230608T120000 SUMMARY;LANGUAGE=en-US:colloq talk by Prof. Bjarne Stroustrup about CS Co lloquia: C++20 – Reaching for the Aims of C++ at 2023-06-08 11:00:00 DESCRIPTION;LANGUAGE=en-US:Out of necessity C++ has been an evolving langu age. 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 programmin g, modularity, the elimination of the preprocessor, and error handling. N aturally, over the years, C++ has acquired many “barnacles” that can becom e obstacles to developing elegant and efficient code. That has been a reco gnized problem since the early days of C – Dennis Ritchie and I talked abo ut 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 directio n. Short bio: Bjarne Stroustrup is the designer and original implement er of C++ as well as the author of The C++ Programming Language (4th Editi on) and A Tour of C++ (2nd edition), Programming: Principles and Practice using C++ (2nd Edition), and many popular and academic publications. Dr . Stroustrup is a Technical Fellow and Managing Director in the technology division of Morgan Stanley in New York City as well as a visiting profess or at Columbia University. He is a member of the US National Academy of En gineering, and an IEEE, ACM, and CHM fellow. He is the recipient of the 20 18 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, sof tware development tools, and programming languages. To make C++ a stable a nd up-to-date base for real-world software development, he has been a lead ing figure with the ISO C++ standards effort for 30 years. He holds a m aster’s in Mathematics from Aarhus University and a PhD in Computer Scienc e from Cambridge University, where he is an honorary fellow of Churchill C ollege. Space is limited - Please register in advance ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il LOCATION:Taub TBD UID:123se24012024104400 END:VEVENT BEGIN:VEVENT DTSTART;TZID="Asia/Jerusalem":20230611T143000 DTEND;TZID="Asia/Jerusalem":20230611T153000 DTSTAMP;TZID="Asia/Jerusalem":20230611T143000 FREEBUSY;FBTYPE=BUSY:20230611T143000/20230611T153000 SUMMARY;LANGUAGE=en-US:Coding_Theory_Semina talk by Prof. Ohad Elishco (Be n-Gurion University) about Coding Theory: Codes Over Absorption Channels at 2023-06-11 14:30:00 DESCRIPTION;LANGUAGE=en-US:In recent years, extensive research has been de dicated 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-mach ines are limited in their capabilities and require communication and netwo rking to tackle complex tasks. By collaborating, these machines can revolu tionize medicine by serving as intelligent drug delivery systems, advanced sensors, and more. The primary challenge in nano-machine communicati on lies in their inability to utilize electromagnetic waves for communicat ion. Hence, an alternative communication method must be employed. When ope rating in living organisms (in-vivo), one potential approach is to leverag e the nervous system, which serves as nature’s communication medium. In this presentation, we introduce a novel communication channel called th e absorption channel, inspired by information transmission through neurons . Our motivation stems from the potential applications of in-vivo nano- an d micro-machines, advancements in medical technology, and brain-machine in terfaces that communicate via the nervous system. We will commence by providing a motivation for the proposed channel. Subsequently, we will pr esent codes capable of correcting absorption errors for any given finite a lphabet. We will explore various scenarios, including single absorption er ror correction over binary alphabets, as well as correction codes for gene ral alphabets. If time permits, we will also delve into multiple absorptio n error-correcting codes over general alphabets. Bio: Ohad Elishco re ceived his B.Sc., M.Sc, and Ph.D degrees in electrical engineering from Be n-Gurion University of the Negev, Israel. Between 2017-2018 he was a postd oc 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. Hi s research interests include constrained coding, Information and coding fo r biology, and dynamical systems. ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il LOCATION:Taub 601 UID:123se24012024104440 END:VEVENT BEGIN:VEVENT DTSTART;TZID="Asia/Jerusalem":20230612T113000 DTEND;TZID="Asia/Jerusalem":20230612T123000 DTSTAMP;TZID="Asia/Jerusalem":20230612T113000 FREEBUSY;FBTYPE=BUSY:20230612T113000/20230612T123000 SUMMARY;LANGUAGE=en-US:msc talk by Guy Horowitz about Causal Strategic Cla ssification at 2023-06-12 11:30:00 DESCRIPTION;LANGUAGE=en-US:When users can benefit from certain predictive outcomes, they may be prone to act to achieve those outcome, e.g., by stra tegically modifying their features. The goal in strategic classification i s therefore to train predictive models that are robust to such behavior. H owever, 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 settin g where true outcomes do change. Focusing on accuracy as our primary objec tive, we show how strategic behavior and causal effects underlie two compl ementing forms of distribution shift. We characterize these shifts, and pr opose 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:123se24012024104370 END:VEVENT BEGIN:VEVENT DTSTART;TZID="Asia/Jerusalem":20230612T143000 DTEND;TZID="Asia/Jerusalem":20230612T153000 DTSTAMP;TZID="Asia/Jerusalem":20230612T143000 FREEBUSY;FBTYPE=BUSY:20230612T143000/20230612T153000 SUMMARY;LANGUAGE=en-US:colloq talk by Avi Wigderson (IAS Princeton) about CS Special Guest Lecture: The Value of Errors in Proofs at 2023-06-12 14:3 0:00 DESCRIPTION;LANGUAGE=en-US:CS Special Guest Lecture by Prof. Avi Wigderson , IAS Princeton, on the Occasion of his Being Awarded an Honorary Doctorat e from the Technion Recently, a group of theoretical computer scientist s 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 "Tsir elson 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. As it happens, both acronyms MIP* and RE represent proof systems, of a very different nat ure. To explain them, we'll take a meandering journey through the classica l and modern definitions of proof. I hope to explain how the methodology o f 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 . The talk does not require special mathematical background. ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il LOCATION:Room 337 taub bld. UID:123se24012024104260 END:VEVENT BEGIN:VEVENT DTSTART;TZID="Asia/Jerusalem":20230613T143000 DTEND;TZID="Asia/Jerusalem":20230613T153000 DTSTAMP;TZID="Asia/Jerusalem":20230613T143000 FREEBUSY;FBTYPE=BUSY:20230613T143000/20230613T153000 SUMMARY;LANGUAGE=en-US:colloq talk by Matan Gavish (The Hebrew University of Jerusalem) about CS Colloquia: Power, Responsibility and Computer Scien ce Training at 2023-06-13 14:30:00 DESCRIPTION;LANGUAGE=en-US:Over the course of four decades, the academic f ield of computer science transformed from a branch of mathematics to a key driver of the evolution of our species. Graduates of academic computer s cience programs today routinely create systems that would have been consid ered, just a century ago, miracles of mythic proportions. Judging by cult ural impact, computer science departments today resemble Hogwarts much mor e than they resemble the theoretical havens they used to be before compute rs got seriously strong. Interestingly, at their core, computer scien ce BSc programs have not changed that much since the 1990's. If indeed wit h great power comes great responsibility, then it is now incumbent on acad emic 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 scienc e literature offers any clues on how to go about this. On the contrary, c omputer science - like all science and technology - detaches technique fro m its human impact, focuses almost exclusively on problem solving, and ten ds to view the world through a narrow quantitative lens. Questions of huma n impact are typically labelled under the obscure term "ethics" and deferr ed wholesale to the social sciences and humanities. Clear basic guideline s for individual choices, analogous to those of the medical and life scien ces, still seem lightyears ahead. I will argue that academic computer s cience departments offer the perfect environment to be asking these questi ons - as Joseph Weizenbaum and Norbert Wiener passionately prophesied at t he 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 H ebrew University, which looked for meaningful ways to form a personal pers pective on the broader implications of digital information technology. Short Bio: Matan Gavish is an Associate Professor in the Hebrew Univ ersity School of Computer Science and Engineering. He is the founder of th e Israel-Singapore center for AI-based urban agriculture (iSURF), and of t he Hebrew University joint CS-Statistics BSc program in Data Science. His research interests include statistical learning, mathematical statistics o f spectral algorithms in high dimensions, applied harmonic analysis, rando m matrix theory, empirical mathematics, reproducible research, data-driven precision agriculture, digital communication studies, and philosophy of t echnology. Matan received the dual B.Sc. degree in Mathematics and Physics from Tel Aviv University in 2006, the M.Sc. degree in Mathematics from th e Hebrew University of Jerusalem in 2008 and the Ph.D. degree in Statistic s from Stanford University in 2014. ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il LOCATION:Room 337 taub bld. UID:123se24012024104350 END:VEVENT BEGIN:VEVENT DTSTART;TZID="Asia/Jerusalem":20230613T160000 DTEND;TZID="Asia/Jerusalem":20230613T170000 DTSTAMP;TZID="Asia/Jerusalem":20230613T160000 FREEBUSY;FBTYPE=BUSY:20230613T160000/20230613T170000 SUMMARY;LANGUAGE=en-US:msc talk by Dan Navon about A Robust Approach to Vi sion-Based Terrain Aided Localization at 2023-06-13 16:00:00 DESCRIPTION;LANGUAGE=en-US:Terrain-aided navigation (TAN) was developed be fore the GPS era to prevent the error growth of inertial navigation. TAN a lgorithms were initially developed to exploit altitude over ground or clea rance measurements from a radar altimeter in combination with a Digital Te rrain 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 t han the original one, as visual observables can only provide a range up to a scale, preventing a straightforward extension of classical TAN techniqu es. The 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 addit ion, a new outlier-rejection step is introduced between the two stages to make the algorithm robust and suitable for real-world data. ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il LOCATION:Zoom Lecture: 94171574353 and Taub 601. UID:123se24012024104310 END:VEVENT BEGIN:VEVENT DTSTART;TZID="Asia/Jerusalem":20230614T123000 DTEND;TZID="Asia/Jerusalem":20230614T133000 DTSTAMP;TZID="Asia/Jerusalem":20230614T123000 FREEBUSY;FBTYPE=BUSY:20230614T123000/20230614T133000 SUMMARY;LANGUAGE=en-US:Theory Seminar talk by Michal Dory (Haifa universit y) about Theory Seminar: Approximate All-Pairs Shortest Paths: Recent Adva nces and Open Questions at 2023-06-14 12:30:00 DESCRIPTION;LANGUAGE=en-US: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^{om ega}) time in unweighted graphs, where omega<2.376 is the exponent of matr ix multiplication. However, these complexities may be high for large graph s, and it is desirable to find faster algorithms. One approach for obtaini ng faster algorithms is to compute approximations for APSP, rather than co mputing the distances exactly. In this talk, I will give an overview of al gorithms for approximate APSP, and discuss recent progress and interesting open questions in this area. The talk will overview some classic results based on [Aingworth, Chekuri, Indyk, and Motwani, 1999] and [Dor, Halperin and Zwick, 2000] as well as recent results based on a joint work with Seb astian Forster, Yasamin Nazari and Tijn de Vos. ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il LOCATION:Taub 201 UID:123se24012024104470 END:VEVENT BEGIN:VEVENT DTSTART;TZID="Asia/Jerusalem":20230614T163000 DTEND;TZID="Asia/Jerusalem":20230614T173000 DTSTAMP;TZID="Asia/Jerusalem":20230614T163000 FREEBUSY;FBTYPE=BUSY:20230614T163000/20230614T173000 SUMMARY;LANGUAGE=en-US:msc talk by Dganit Hanania about On the Capacity of DNA Labeling at 2023-06-14 16:30:00 DESCRIPTION;LANGUAGE=en-US:DNA labeling is a powerful tool in molecular bi ology and biotechnology that allows for the visualization, detection, and study of DNA at the molecular level. Under this paradigm, a DNA molecule i s being labeled by specific k patterns and is then imaged. Then, the resul ted image is modeled as a (k + 1)-ary sequence in which any non-zero symbo l indicates on the appearance of the corresponding label in the DNA molecu le. 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 thi s labeling process. The labeling capacity is computed for any single label and several results are provided for multiple labels as well. Moreover, w e provide the optimal minimal number of labels of length one or two that a re needed in order to gain labeling capacity of 2. ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il LOCATION:Taub 601 UID:123se24012024104170 END:VEVENT BEGIN:VEVENT DTSTART;TZID="Asia/Jerusalem":20230618T143000 DTEND;TZID="Asia/Jerusalem":20230618T153000 DTSTAMP;TZID="Asia/Jerusalem":20230618T143000 FREEBUSY;FBTYPE=BUSY:20230618T143000/20230618T153000 SUMMARY;LANGUAGE=en-US:Coding_Theory_Semina talk by Yonatan Yehezkeally (U niversity of Munich) about Coding Theory: Resilient Repeat-free Codes at 2 023-06-18 14:30:00 DESCRIPTION;LANGUAGE=en-US:Repeat-free codes are used to ensure unique rec onstruction from fragmentation, assuming full (uniform) read-coverage of s ubstrings, with applications to DNA-based storage systems. In this talk, w e explore a generalization aimed at resilience to pre-fragmentation noise, and study existence results as well as explicit constructions. Yonatan Yehezkeally is the Carl Friedrich von Siemens post-doctoral research fell ow of the Alexander von Humboldt Foundation, in the Associate Professorshi p of Coding and Cryptography (Prof. Wachter-Zeh), School of Computation, I nformation and Technology, Technical University of Munich. His research in terests include coding for novel storage media, with a focus on DNA-based storage and nascent sequencing technologies, as well as combinatorial stru ctures and finite group theory. Yonatan received the Ph.D. degree in El ectrical and Computer Engineering in 2020, from Ben-Gurion University of t he Negev, Beer-Sheva, Israel. Before that, he received the B.Sc. degree (c um laude) in Mathematics and the M.Sc. degree (summa cum laude) in Electri cal and Computer Engineering, in 2013 and 2017 respectively, also from Ben -Gurion University of the Negev. ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il LOCATION:Taub 601 UID:123se24012024104480 END:VEVENT BEGIN:VEVENT DTSTART;TZID="Asia/Jerusalem":20230620T143000 DTEND;TZID="Asia/Jerusalem":20230620T153000 DTSTAMP;TZID="Asia/Jerusalem":20230620T143000 FREEBUSY;FBTYPE=BUSY:20230620T143000/20230620T153000 SUMMARY;LANGUAGE=en-US:colloq talk by Prof. Adi Shamir (Weizmann Institute of Science) about CS Colloquia: Facial Misrecognition Systems at 2023-06 -20 14:30:00 DESCRIPTION;LANGUAGE=en-US:In this talk I will describe how to plant novel types of backdoors in any facial recognition model based on the popular a rchitecture 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 specifi c 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 an d decide that they represent different persons (an anonymity attack), or t ake any two images of a particular pair of persons and decide that they re present the same person (a confusion attack), with almost no effect on the correctness of its decisions for other persons. Uniquely, we show that mu ltiple backdoors can be independently installed by multiple attackers who may not be aware of each other's existence with almost no interference. Joint work with Irad Zehavi.Based on joint works with Venkat Guru swami and Peter Manohar.
ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il LOCATION:Taub 201 UID:123se24012024104630 END:VEVENT BEGIN:VEVENT DTSTART;TZID="Asia/Jerusalem":20230709T110000 DTEND;TZID="Asia/Jerusalem":20230709T120000 DTSTAMP;TZID="Asia/Jerusalem":20230709T110000 FREEBUSY;FBTYPE=BUSY:20230709T110000/20230709T120000 SUMMARY;LANGUAGE=en-US:phd talk by Alona Levy about Deep Learning and St atistical Methods for Digital Pathology and Molecular Measurements at 2023 -07-09 11:00:00 DESCRIPTION;LANGUAGE=en-US:Digital analysis of pathology whole-slide image s is fast becoming a game changer in cancer diagnosis and treatment. Speci fically, deep learning methods have shown great potential to support patho logy analysis, with recent studies identifying molecular traits that were not previously recognized in pathology H&E whole-slide images. Simultaneou s to these developments, it is becoming increasingly evident that tumor he terogeneity is an important determinant of cancer prognosis and susceptibi lity to treatment, and should therefore play a role in the evolving practi ces of matching treatment protocols to patients. In this talk, I will pres ent our work on spatially resolving bulk mRNA and miRNA expression levels on pathology whole-slide images (WSIs). I will further present a statistic al method we developed to spatially characterize tumor heterogeneity from the inferred gene expression levels and demonstrate that it applies to a w ide variety of spatial data, including 3D data like brain MRI scans. Final ly, I will present a deep learning model we developed to predict DNA methy lation levels. ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il LOCATION:Zoom Lecture: 93993434972 and Taub 601 UID:123se24012024104620 END:VEVENT BEGIN:VEVENT DTSTART;TZID="Asia/Jerusalem":20230711T113000 DTEND;TZID="Asia/Jerusalem":20230711T123000 DTSTAMP;TZID="Asia/Jerusalem":20230711T113000 FREEBUSY;FBTYPE=BUSY:20230711T113000/20230711T123000 SUMMARY;LANGUAGE=en-US:phd talk by Bahjat Kawar about Diffusion Models for Image Restoration at 2023-07-11 11:30:00 DESCRIPTION;LANGUAGE=en-US:Denoising Diffusion Probabilistic Models (DDPM) , also known as diffusion models, have recently emerged as state-of-the-ar t generative models, synthesizing images with unprecedented quality and re alism. At their core, diffusion models employ an MSE-trained denoiser neur al network in an iterative scheme, transforming random noise into pristine images. Theoretically, this algorithm is proven to draw samples from a le arned prior image distribution. In our work, we adapt pre-trained diffusio n models for the task of image restoration, mainly focusing on linear inve rse problems. We present a novel outlook on inverse problem solving, posin g it as a posterior sampling task rather than an optimization problem. Thi s approach introduces several advantages such as improved perceptual quali ty, multiple solutions, and uncertainty quantification. Moreover, our meth od does not require task-specific training, and we demonstrate its use for inpainting, super resolution, deblurring, colorization, and more. ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il LOCATION:Zoom Lecture: 96270781265 and Taub 401 UID:123se24012024104460 END:VEVENT BEGIN:VEVENT DTSTART;TZID="Asia/Jerusalem":20230716T100000 DTEND;TZID="Asia/Jerusalem":20230716T110000 DTSTAMP;TZID="Asia/Jerusalem":20230716T100000 FREEBUSY;FBTYPE=BUSY:20230716T100000/20230716T110000 SUMMARY;LANGUAGE=en-US:msc talk by Liran Farhi about Movement as a Langu age: Unleashing the Power of Indoor Movement Analysis for Semantic Place P rediction at 2023-07-16 10:00:00 DESCRIPTION;LANGUAGE=en-US:The proliferation of modern mobile phones has o pened up unprecedented opportunities for leveraging location-tracking capa bilities to extract individual mobility patterns and contextual informatio n. However, existing approaches heavily rely on analyzing geolocation data obtained from Global Navigation Satellite System (GNSS) observations, whi ch are limited when used in enclosed spaces. This talk presents new algori thms for efficient mobility data analysis and contextual learning by utili zing the WiFi infrastructure that exists today in most indoor environments . Firstly, I will introduce SWATSON, an unsupervised trajectory segmentati on algorithm that is designed to partition a continuous sequence of data p oints into homogeneous segments. SWATSON effectively utilizes temporal con straints, mitigates noise, and exhibits applicability across various domai ns. By employing SWATSON, I will demonstrate an approach to analyzing move ment in indoor environments without needing a localization process. Conseq uently, the integration of SWATSON will be explored to create personalized semantic categorical place labeling, such as assigning the label "work" t o specific locations. I will present a supervised learning-based classific ation 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-ba sed 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:123se24012024104650 END:VEVENT BEGIN:VEVENT DTSTART;TZID="Asia/Jerusalem":20230723T133000 DTEND;TZID="Asia/Jerusalem":20230723T143000 DTSTAMP;TZID="Asia/Jerusalem":20230723T133000 FREEBUSY;FBTYPE=BUSY:20230723T133000/20230723T143000 SUMMARY;LANGUAGE=en-US:ceClub talk by Joseph Friedman (University of Texas at Dallas) about ceClub: Reversible, Neuromorphic, Reservoir, and Secure Computing with Spintronic Phenomena at 2023-07-23 13:30:00 DESCRIPTION;LANGUAGE=en-US:The rich physics present in a wide range of spi ntronic materials and devices provide opportunities for a variety of compu ting applications. This presentation will describe six distinct proposals to leverage spintronic phenomena for reversible computing, neuromorphic co mputing, reservoir computing, and hardware security. The presentation will begin with a solution for reversible computing in which magnetic skyrmion s propagate and interact in a scalable system with the potential for energ y dissipation below the Landauer limit, followed by a paradigm for operati ng Boolean logic at terahertz clock frequencies utilizing the magnetoresis tance of low-dimensional materials. Three neuromorphic systems for emulati ng neurobiological behavior with spintronic phenomena will then be present ed: a purely-spintronic system that enables unsupervised learning with mag netic domain wall neurons and synapses, a reservoir computing system based on the dynamics of frustrated nanomagnets, and an approach for unsupervis ed learning that marks the first experimental demonstration of a neuromorp hic network directly implemented with MTJ synapses. This presentation will conclude with a logic locking paradigm based on nanomagnet logic, the fir st logic locking system that is secure against both physical and algorithm ic attacks. Bio: Dr. Joseph S. Friedman is an associate professor of E lectrical & 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 undergr aduate degrees from Dartmouth College. He was previously a CNRS Research A ssociate with Université Paris-Saclay, a Summer Faculty Fellow at the U.S. Air Force Research Laboratory, a Visiting Professor at Politecnico di Tor ino, a Guest Scientist at RWTH Aachen University, and worked on logic desi gn automation at Intel Corporation. Dr. Friedman is a member of the edi torial boards of Scientific Reports and IEEE Transactions on Nanotechnolog y, and previously the Microelectronics Journal. He is a conference chair o f SPIE Spintronics, has served on numerous conference technical program co mmittees, and is the founder and chairperson of the Texas Symposium on Com puting with Emerging Technologies (ComET). He has also been awarded the Na tional Science Foundation (NSF) Faculty Early Career Development Program ( CAREER) Award. ATTENDEE;CUTYPE=GROUP;PARTSTAT=TENTATIVE:mailto:webmaster@cs.technion.ac.il LOCATION:Room 861, EE Meyer Building & Zoom Lecture: 94673013539 UID:123se24012024104660 END:VEVENT BEGIN:VEVENT DTSTART;TZID="Asia/Jerusalem":20230725T113000 DTEND;TZID="Asia/Jerusalem":20230725T123000 DTSTAMP;TZID="Asia/Jerusalem":20230725T113000 FREEBUSY;FBTYPE=BUSY:20230725T113000/20230725T123000 SUMMARY;LANGUAGE=en-US:phd talk by Roee Francos about Multi-Agent Teamwork in Search for Smart Opponents Detection at 2023-07-25 11:30:00 DESCRIPTION;LANGUAGE=en-US:Cooperative Multi-Agent teams can be deployed i n many interesting and important domains such as industry, transportation, agriculture, security and more. In this talk, I will introduce key result s from my research, primarily focusing on theoretical work concerned with search for smart agents by UAV teams. Suppose that in a given planar circu lar 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 fo r the sweeping team of agents, and present guaranteed search techniques fo r single agent and multi agent teams. These search procedures enable both confinement of the smart agents to their original domain as well as comple te detection of all of them by searching the entire expanding domain. Furt hermore, we investigate the dual problem of devising guaranteed defense po licies for protecting a given region from the entrance of smart mobile age nts by detecting them using a team of sweeping agents. The desired outcome of the developed protocols is a defense strategy of the original domain a nd for its expansion.To particip
ate, register at the link
In the program:
Dr. Inbal Tsafir-Lavia, CE
O and co-entrepreneur at Nevia Bio, which is developing a test for the ear
ly diagnosis of ovarian cancer
Michal Shanhav, Talent Acquisition P
artner, IBM Research, HR Israel
To the community page on Facebook