To join the email distribution list of the cs colloquia, please visit the list subscription page.
Computer Science events calendar in HTTP ICS format for of Google calendars, and for Outlook.
Academic Calendar at Technion site.
Diffusion models have emerged as the leading approach for high-quality image synthesis and demonstrate exceptional versatility in solving inverse problems through their powerful learned image priors. In this seminar, we explore how these generative priors enable adaptive compressed sensing for real-world active acquisition applications, including MRI and CT imaging, where intelligent measurement selection can dramatically reduce scan times while preserving reconstruction quality.
We further demonstrate how these same principles extend naturally to image compression, leveraging the diffusion prior to achieve efficient encoding and high-fidelity reconstruction.
Motivated by limitations in existing posterior sampling methods, we introduce a novel model architecture specifically designed for inverse problems that is both theoretically justified and computationally efficient. Collectively, these contributions establish a unified framework for deploying diffusion models across medical imaging, image compression, and image restoration, advancing both the practical applicability and theoretical foundations of generative models for inverse problems.
Noam Elata is a Ph.D. candidate under the supervision of Prof. Michael Elad and Prof. Tomer Michaeli.
Taub 337
The field of robotics is wildly exciting and rapidly gaining worldwide attention, yet it is often an enigma in terms of its scope and scientific foundations. Robotics involves the design, programming, and analysis of movable machines that accomplish useful work through sensing and manipulation of the surrounding world. Throughout the decades it has been varyingly viewed as an application field of more mature disciplines such as computer science (AI, algorithms, machine learning) and mechanical engineering (kinematics, dynamics, nonlinear control). This talk will argue that robotics has its own unique and growing scientific core, with deep questions and modelling challenges that should inspire new directions in computer science, engineering, and even pure mathematics.
We will start with a Turing-inspired way to view robotics or embodied AI, leading to some of our recent results that characterize minimally sufficient amounts of sensing, actuation, or computation that are required to solve physical tasks. Questions addressed include: How are learning, planning, and control related? How do we know when it is impossible to solve a task? When will learning fail, even with an infinite amount of data? Does a universal action sequence exist that would cause a robot to solve any possible task without modification? How important are semantics and representations? Interspersed throughout the talk will be results and perspective from my research in the field over three decades, produced with many inspiring students, mentors, and collaborators.
Bio: Steven M. LaValle has been Professor of Computer Science and Engineering, in Robotics and Virtual Reality, at the University of Oulu, Finland since 2018. Since 2001, he has been a professor in the Department of Computer Science at the University of Illinois. He has also held positions at Stanford University and Iowa State University. His research interests include robotics, virtual reality, sensor fusion, planning algorithms, computational geometry, and control theory. In research, he is mostly known for his introduction of the Rapidly exploring Random Tree (RRT) algorithm, which is widely used in robotics and other engineering fields. He also authored the books Planning Algorithms, Sensing and Filtering, and Virtual Reality. He currently leads an Advanced Grant project from the European Research Council on the Foundations of Perception Engineering.
With regard to industry, he was an early founder and chief scientist of Oculus VR, acquired by Facebook for $3 billion in 2014, where he developed patented tracking technology for consumer virtual reality and led a team of perceptual psychologists to provide principled
approaches to virtual reality system calibration, health and safety, and the design of comfortable user experiences. From 2016 to 2017, he was a Vice President and Chief Scientist of VR/AR/MR at Huawei Technologies, where he was a leader in mobile product development on a global scale. He has worked as an angel investor and adviser to startups in robotics and virtual reality.
Technion Host: Oren Salzman
Modern machine learning models often struggle to remain reliable under challenging conditions such as distribution shifts, adversarial perturbations, or limited data. My research focuses on improving robustness and generalization by leveraging task-specific structure at inference time, without requiring additional training or data. I present methods that adapt either the inputs or the context of pretrained models to better align with the underlying task. In the visual domain, these approaches enhance robustness by transforming or projecting inputs toward meaningful class- or data-manifold representations. In the language domain, they refine prompt representations to extract more effective information from few-shot examples. Together, these contributions demonstrate a unified principle: carefully adapting representations to reflect task-relevant structure can substantially improve the reliability and generalization of modern machine learning systems across both vision and language.