Provably Learning Diverse Features in Multi-View Data with Midpoint Mixup

TL;DR

This paper provides a theoretical explanation for Mixup's success in learning multiple features in multi-view data, showing it enables neural networks to learn diverse features that standard training often misses.

Contribution

It demonstrates that Mixup can theoretically and empirically facilitate learning multiple features per class, unlike traditional empirical risk minimization.

Findings

Mixup enables learning multiple features per class in theory.

Standard ERM tends to learn only one feature per class.

Empirical results on image benchmarks support the theoretical insights.

Abstract

Mixup is a data augmentation technique that relies on training using random convex combinations of data points and their labels. In recent years, Mixup has become a standard primitive used in the training of state-of-the-art image classification models due to its demonstrated benefits over empirical risk minimization with regards to generalization and robustness. In this work, we try to explain some of this success from a feature learning perspective. We focus our attention on classification problems in which each class may have multiple associated features (or views) that can be used to predict the class correctly. Our main theoretical results demonstrate that, for a non-trivial class of data distributions with two features per class, training a 2-layer convolutional network using empirical risk minimization can lead to learning only one feature for almost all classes while training with a specific instantiation of Mixup succeeds in learning both features for every class. We also show empirically that these theoretical insights extend to the practical settings of image benchmarks modified to have multiple features.