Memory-efficient Learning for Large-scale Computational Imaging -- NeurIPS deep inverse workshop

TL;DR

This paper introduces a memory-efficient learning method for large-scale computational imaging systems, leveraging network reversibility to enable deep neural network training without excessive memory use.

Contribution

It proposes a reversible network-based training approach that overcomes GPU memory limitations in large-scale imaging applications.

Findings

Successfully applied to super-resolution microscopy

Effective in multi-channel MRI reconstruction

Enables scalable deep learning for large imaging systems

Abstract

Computational imaging systems jointly design computation and hardware to retrieve information which is not traditionally accessible with standard imaging systems. Recently, critical aspects such as experimental design and image priors are optimized through deep neural networks formed by the unrolled iterations of classical physics-based reconstructions (termed physics-based networks). However, for real-world large-scale systems, computing gradients via backpropagation restricts learning due to memory limitations of graphical processing units. In this work, we propose a memory-efficient learning procedure that exploits the reversibility of the network's layers to enable data-driven design for large-scale computational imaging. We demonstrate our methods practicality on two large-scale systems: super-resolution optical microscopy and multi-channel magnetic resonance imaging.