[ICASSP 2023] Deep Proximal Gradient Method for Learned Convex Regularizers скачать в хорошем качестве

[ICASSP 2023] Deep Proximal Gradient Method for Learned Convex Regularizers

MERL Researcher Hassan Mansour presents his paper tilted "Deep Proximal Gradient Method for Learned Convex Regularizers" for the IEEE International Conference on Acoustics, Speech, and Signal Processing (ICASSP) held June 4 - 10, 2023. The paper was co-authored with MERL researchers Yanting Ma, Perry Wang, and Petros Boufounos, and with former intern Aaron Berk from the Department of Mathematics and Statistics, McGill University. Paper: https://ieeexplore.ieee.org/document/..., https://www.merl.com/publications/doc... Abstract: We consider the problem of simultaneously learning a convex penalty function and its proximity operator for image reconstruction from incomplete measurements. Our goal is to apply Accelerated Proximal Gradient Method (APGM) using a learned proximity operator in place of the true proximity operator of the learned penalty function. Starting from a Gaussian image denoiser, we learn an associated penalty function and its proximity operator. The learned penalty function offers provable reconstruction guarantees, whereas access to its proximity operator presents the opportunity to achieve APGM convergence rates, which are faster than those of subgradient descent approaches.