mixup: Beyond Empirical Risk Minimization

TL;DR

Mixup is a simple data augmentation technique that trains neural networks on convex combinations of examples to improve generalization, robustness, and training stability across various datasets and models.

Contribution

This paper introduces mixup, a novel regularization method that enhances neural network performance by interpolating training data and labels, reducing memorization and adversarial vulnerability.

Findings

Improves generalization on multiple datasets

Reduces memorization of corrupt labels

Increases robustness to adversarial examples

Abstract

Large deep neural networks are powerful, but exhibit undesirable behaviors such as memorization and sensitivity to adversarial examples. In this work, we propose mixup, a simple learning principle to alleviate these issues. In essence, mixup trains a neural network on convex combinations of pairs of examples and their labels. By doing so, mixup regularizes the neural network to favor simple linear behavior in-between training examples. Our experiments on the ImageNet-2012, CIFAR-10, CIFAR-100, Google commands and UCI datasets show that mixup improves the generalization of state-of-the-art neural network architectures. We also find that mixup reduces the memorization of corrupt labels, increases the robustness to adversarial examples, and stabilizes the training of generative adversarial networks.