Manifold Mixup: Better Representations by Interpolating Hidden States

TL;DR

Manifold Mixup is a regularization technique that interpolates hidden states in neural networks to improve generalization, robustness, and decision boundary smoothness without significant computational overhead.

Contribution

It introduces Manifold Mixup, a novel regularizer that interpolates hidden representations to enhance neural network robustness and generalization, supported by theoretical and empirical validation.

Findings

Improves robustness to adversarial attacks

Enhances test log-likelihood performance

Produces smoother decision boundaries

Abstract

Deep neural networks excel at learning the training data, but often provide incorrect and confident predictions when evaluated on slightly different test examples. This includes distribution shifts, outliers, and adversarial examples. To address these issues, we propose Manifold Mixup, a simple regularizer that encourages neural networks to predict less confidently on interpolations of hidden representations. Manifold Mixup leverages semantic interpolations as additional training signal, obtaining neural networks with smoother decision boundaries at multiple levels of representation. As a result, neural networks trained with Manifold Mixup learn class-representations with fewer directions of variance. We prove theory on why this flattening happens under ideal conditions, validate it on practical situations, and connect it to previous works on information theory and generalization. In spite of incurring no significant computation and being implemented in a few lines of code, Manifold Mixup improves strong baselines in supervised learning, robustness to single-step adversarial attacks, and test log-likelihood.