Invert to Learn to Invert
This addresses memory limitations for researchers and practitioners in medical imaging and inverse problems, though it is incremental as it builds on existing iterative learning methods.
The authors tackled the memory bottleneck in training deep iterative models for inverse problems by proposing an invertible network architecture that eliminates the need to store intermediate activations, enabling training of 400-layer models on 3D MRI data and achieving state-of-the-art reconstruction performance.
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.