# How Does Mixup Help With Robustness and Generalization?

Source: [How Does Mixup Help With Robustness and Generalization?](<https://www.sota2.com/research/paper/68fa69951e209131cea22d47>)

Collection index: [llms.txt](<https://www.sota2.com/research/llms.txt>)

Data updated: 2025-10-25T23:03:46.301Z

Authors: Linjun Zhang, Zhun Deng, Kenji Kawaguchi, Amirata Ghorbani, James Zou

Published: 2020-10-09T21:38:14.000Z

## Abstract

Mixup is a popular data augmentation technique based on taking convex combinations of pairs of examples and their labels. This simple technique has been shown to substantially improve both the robustness and the generalization of the trained model. However, it is not well-understood why such improvement occurs. In this paper, we provide theoretical analysis to demonstrate how using Mixup in training helps model robustness and generalization. For robustness, we show that minimizing the Mixup loss corresponds to approximately minimizing an upper bound of the adversarial loss. This explains why models obtained by Mixup training exhibits robustness to several kinds of adversarial attacks such as Fast Gradient Sign Method (FGSM). For generalization, we prove that Mixup augmentation corresponds to a specific type of data-adaptive regularization which reduces overfitting. Our analysis provides new insights and a framework to understand Mixup.

## Recorded benchmark results

A research result applies to its stated dataset, split, metric and evaluation configuration. Compare like-for-like settings, preserve the metric's unit and direction, and distinguish paper-reported results from independent reproduction. The best result recorded here is not a claim of exhaustive, current global coverage.

- [Image Classification Calibration on Tiny ImageNet C](<https://www.sota2.com/research/sota/image-classification-calibration-on-tiny-imagenet-c.md>): Image Classification Calibration; Tiny-ImageNet-C; Accuracy: 26.26.

- [Image Classification Calibration on CIFAR-10-C](<https://www.sota2.com/research/sota/image-classification-calibration-on-cifar-10-c.md>): Image Classification Calibration; CIFAR-10-C; Accuracy: 75.41.

- [Image Classification Calibration on CIFAR-100-C](<https://www.sota2.com/research/sota/image-classification-calibration-on-cifar-100-c.md>): Image Classification Calibration; CIFAR-100-C; Accuracy: 53.38.

- [Image Classification Calibration on Camelyon 17](<https://www.sota2.com/research/sota/image-classification-calibration-on-camelyon-17.md>): Image Classification Calibration; CAMELYON 17; Accuracy: 88.33.

- [Image Classification Calibration on iWildCam](<https://www.sota2.com/research/sota/image-classification-calibration-on-iwildcam.md>): Image Classification Calibration; iWILDCam; Accuracy: 75.85.

- [Image Classification Calibration on FMoW](<https://www.sota2.com/research/sota/image-classification-calibration-on-fmow.md>): Image Classification Calibration; fMoW; Accuracy: 51.53.

## Original sources and implementation

[Original paper](<https://arxiv.org/abs/2010.04819>)

[PDF](<https://arxiv.org/pdf/2010.04819.pdf>)