?Generativemodels and Unsupervised methods for Inverse problems?
Alex Dimakis, PhD
Electricaland Computer Engineering Department
Universityof Texas at Austin
Abstract: Moderndeep generative modelslike GANs, VAEs, invertible flows and Score-based models are demonstratingexcellent performance in representing high-dimensional distributions,especially for images. We will show how they can be usedto solve inverseproblems like denoising, filling missing data, and recovery from linear projections.We generalize compressed sensing theory beyond sparsity, extending RestrictedIsometries to sets created by deep generative models. Our recent resultsinclude establishing theoretical results for Langevin sampling fromfull-dimensional generative models and fairness guarantees for inverseproblems.
Biography: Alex Dimakis is a Professor at the ECEdepartment at UT Austin and the co-director of the National AI Institute on theFoundations of Machine Learning (IFML). He received his Ph.D. from UC Berkeleyand the Diploma degree from the National Technical University of Athens. Hereceived several awards including the James Massey Award, NSF Career, a Googleresearch award, the Eli Jury dissertation award and the 2012 joint InformationTheory and Communications Society Best Paper Award. His research interestsinclude information theory, coding theory and machine learning.