Unfolded proximal neural networks for computational imaging

Title

Unfolded proximal neural networks for computational imaging

Speaker

Audrey Repetti - Heriot Watt University, Edinburgh, UK

Abstract

A common approach to solve inverse imaging problems relies on finding a maximum a posteriori (MAP) estimate of the original unknown image, by solving a minimization problem. In this context, iterative proximal algorithms are widely used, enabling to handle non-smooth functions and linear operators. Recently, these algorithms have been paired with deep learning strategies, to further improve the estimate quality. In particular, proximal neural networks (PNNs) have been introduced, obtained by unrolling a proximal algorithm as for finding a MAP estimate, but over a fixed number of iterations, with learned linear operators and parameters. As PNNs are based on optimization theory, they are very flexible, and can be adapted to any image restoration task, as soon as a proximal algorithm can solve it. They further have much lighter architectures than traditional networks.

In this presentation, we describe a unified framework to build PNNs for the Gaussian denoising task, based on both the dual-FB and the primal-dual Chambolle-Pock algorithms. We further show that accelerated inertial versions of these algorithms enable skip connections in the associated NN layers. We propose different learning strategies for our PNN framework, and investigate their robustness (Lipschitz property) and denoising efficiency. Finally, we show that the proposed PNNs can be either plugged in a forward-backward algorithm for an image deblurring problem, or used for edge detection.

Bio

Audrey Repetti is an Associate Professor at Heriot-Watt University, affiliated with the school of Mathematical and Computer Sciences. Her research encompasses optimization and deep learning theories, inverse problems, with application to computational imaging. She was awarded in 2022 a Fellowship from the Royal Society of Edinburgh, for her works on optimisation for data science. She was an AE for IEEE SPL (2022-2024) and DSP (2021-2024). She is an AE for IEEE TCI since 2025; elected member of the IEEE SPTM TC since 2024 and EURASIP TMTSP TAC since 2022; and organizing committee member of the SIAM Conference on Imaging Sciences 2022 & 2024, IEEE MLSP conference 2024, IEEE SSP conference 2025, SIAM Conference on Optimization 2026.

When

Thursday May 8th, 12:00

Where

Room 322, UniGe DIBRIS/DIMA, Via Dodecaneso 35