Reversible Architectures for Arbitrarily Deep Residual Neural Networks

TL;DR

This paper introduces reversible neural network architectures inspired by differential equations, enabling arbitrarily deep, memory-efficient, and stable networks that achieve state-of-the-art results on image classification tasks.

Contribution

The paper develops a theoretical framework for reversible neural networks based on ODE interpretation, leading to new architectures that are deep, stable, and memory-efficient.

Findings

Achieved state-of-the-art or competitive results on CIFAR-10, CIFAR-100, and STL-10 datasets.

Demonstrated memory efficiency and stability in training very deep networks.

Showed improved performance with fewer training data.

Abstract

Recently, deep residual networks have been successfully applied in many computer vision and natural language processing tasks, pushing the state-of-the-art performance with deeper and wider architectures. In this work, we interpret deep residual networks as ordinary differential equations (ODEs), which have long been studied in mathematics and physics with rich theoretical and empirical success. From this interpretation, we develop a theoretical framework on stability and reversibility of deep neural networks, and derive three reversible neural network architectures that can go arbitrarily deep in theory. The reversibility property allows a memory-efficient implementation, which does not need to store the activations for most hidden layers. Together with the stability of our architectures, this enables training deeper networks using only modest computational resources. We provide both theoretical analyses and empirical results. Experimental results demonstrate the efficacy of our architectures against several strong baselines on CIFAR-10, CIFAR-100 and STL-10 with superior or on-par state-of-the-art performance. Furthermore, we show our architectures yield superior results when trained using fewer training data.