Invertible Residual Networks

TL;DR

This paper demonstrates that standard ResNet architectures can be made invertible with a simple normalization step, enabling their use for classification, density estimation, and generation without architectural restrictions.

Contribution

The authors introduce a method to make ResNets invertible using a normalization step, allowing a single model to perform multiple tasks including generative modeling.

Findings

Invertible ResNets perform competitively with state-of-the-art classifiers.

They can be used for density estimation and generation.

The approach requires only a simple normalization during training.

Abstract

We show that standard ResNet architectures can be made invertible, allowing the same model to be used for classification, density estimation, and generation. Typically, enforcing invertibility requires partitioning dimensions or restricting network architectures. In contrast, our approach only requires adding a simple normalization step during training, already available in standard frameworks. Invertible ResNets define a generative model which can be trained by maximum likelihood on unlabeled data. To compute likelihoods, we introduce a tractable approximation to the Jacobian log-determinant of a residual block. Our empirical evaluation shows that invertible ResNets perform competitively with both state-of-the-art image classifiers and flow-based generative models, something that has not been previously achieved with a single architecture.