Invert to Learn to Invert

TL;DR

This paper introduces an invertible network-based iterative inverse model that achieves constant memory usage, enabling training of very deep models for 3D MRI reconstruction, resulting in state-of-the-art performance.

Contribution

The authors propose a novel invertible network approach for inverse problems that significantly reduces memory requirements, allowing for deeper and more expressive models.

Findings

Achieved state-of-the-art MRI reconstruction results.

Enabled training of 400-layer models on 3D volumes.

Demonstrated constant memory usage during training.

Abstract

Iterative learning to infer approaches have become popular solvers for inverse problems. However, their memory requirements during training grow linearly with model depth, limiting in practice model expressiveness. In this work, we propose an iterative inverse model with constant memory that relies on invertible networks to avoid storing intermediate activations. As a result, the proposed approach allows us to train models with 400 layers on 3D volumes in an MRI image reconstruction task. In experiments on a public data set, we demonstrate that these deeper, and thus more expressive, networks perform state-of-the-art image reconstruction.