Learned Proximal Operators: from Denoising to Sampling
About this Topic:
Proximal operators are central to signal processing, imaging, and machine learning, forming the backbone of many foundational optimization algorithms. Their practical utility stems from the availability of simple, structured functions that admit efficient proximal mappings—yet this requirement often restricts them to analytically defined forms. Then, how can we design functions that remain simple and proximable, while being learned directly from data?
In this talk, the presenter will introduce learned proximal networks: parametric function classes that provably provide proximal operators of data log-densities. These constructions unify learning and optimization, yielding principled, data-driven proximal mappings. We will then see how they naturally provide new denoising models, act as core primitives for solving inverse problems, and enable proximal variants of diffusion models with improved convergence properties.
About the Presenter:
Jeremias Sulam received the bioengineering degree from Universidad Nacional de Entre Ríos, Argentina, and the Ph.D. in computer science from the Technion – Israel Institute of Technology, in 2013 & 2018 respectively.
He is currently the William R. Brody Faculty Scholar and Associate Professor of Biomedical Engineering at Johns Hopkins University. He is a core faculty member of the Mathematical Institute for Data Science (MINDS), the Center for Imaging Science, the Kavli Neuroscience Discovery Institute, and the Data Science and AI Institute. His research lies at the intersection of machine learning theory and biomedical imaging, with a focus on parsimonious data representations, robustness, and the interpretability and auditing of data-driven models.
Dr. Sulam is a recipient of the Best Graduates Award from the Argentine National Academy of Engineering and the NSF CAREER Award. His work is motivated by challenges in diagnostic imaging, inverse problems, and biomarker discovery, with applications spanning radiology, neuroscience, and digital pathology.
Want to learn more about upcoming events & webinars?
Visit the events section of the Signal Processing website to see all upcoming lectures, workshops, webinars, and more.