Learned Regularizers for Image Reconstruction

by Stanislas Ducotterd (EPFL)

Thursday 23 Apr 2026, 11:30 → 13:00 Europe/Zurich

OHSA/B17

Abstract:
This presentation explores the evolution of image reconstruction, beginning with an intuitive overview of classical regularizers and their role in solving inverse problems. It then examines modern deep-learning-driven iterative techniques, which, despite their powerful denoising capabilities, can struggle with instability and fail to converge. Finally, the talk surveys the current state of research dedicated to overcoming this challenge, illustrating how the field is working to restore rigorous mathematical convergence guarantees without sacrificing the empirical performance of state-of-the-art neural networks.

TEAMS link