Noam Elata (Technion): ” Diffusion Models for Posterior Sampling and Adaptive Sensing”

Speaker: Noam Elata (Technion)

Title: “Diffusion Models for Posterior Sampling and Adaptive Sensing“

Video: Click here to view

Abstract: 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 several recent works, 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.

Bio:Noam Elata is a PhD candidate at the Technion, advised by Prof. Michael Elad (CS) and Prof. Tomer Michaeli (ECE). His research focuses on sampling and training algorithms for diffusion models, with emphasis on inverse problems and posterior sampling. His broader interests span theoretical aspects of deep learning, novel neural architectures, and applications of LLMs and VLMs. Currently, he is also developing and accelerating generative video models at Apple.