mlsanityData to 7 Oct 2026

How Does Mixup Help With Robustness and Generalization?

arXiv 2010.04819 ↗First posted 9 Oct 2020cs.LG stat.MLcited by 193

Not on the map. Why?

Abstract (arXiv)

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.

Key facts. Cited by 193 papers in the map (data through 7 Oct 2026). First posted on arXiv 9 Oct 2020.

Builds on

No reference list yet.

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2021: 30; 2022: 52; 2023: 51; 2024: 30; 2025: 25; 2026: 5.

193 papers.

Sort:

By area:

  1. 2022Transformers Learn Shortcuts to Automatacited by 260
  2. 2022Improving Out-of-Distribution Robustness via Selective Augmentationcited by 202
  3. 2023Diverse Data Augmentation with Diffusions for Effective Test-time Prompt Tuningcited by 150
  4. 2022G-Mixup: Graph Data Augmentation for Graph Classificationcited by 145
  5. 2021Disparities in Dermatology AI: Assessments Using Diverse Clinical Imagescited by 132
  6. 2022Out-Of-Distribution Generalization on Graphs: A Surveycited by 97
  7. 2021Representational Continuity for Unsupervised Continual Learningcited by 95
  8. 2022Complete Verification via Multi-Neuron Relaxation Guided Branch-and-Boundcited by 92
  9. 2023How Does Information Bottleneck Help Deep Learning?cited by 92
  10. 2021An Unconstrained Layer-Peeled Perspective on Neural Collapsecited by 87
  11. 2023Long-tailed Visual Recognition via Gaussian Clouded Logit Adjustmentcited by 75
  12. 2023Discover and Cure: Concept-aware Mitigation of Spurious Correlationcited by 71
38 more
  1. 2021Editing a classifier by rewriting its prediction rulescited by 64
  2. 2022Expanding Small-Scale Datasets with Guided Imaginationcited by 61
  3. 2021When and How Mixup Improves Calibrationcited by 59
  4. 2022C-Mixup: Improving Generalization in Regressioncited by 57
  5. 2021Unleashing the Power of Contrastive Self-Supervised Visual Models via Contrast-Regularized Fine-Tuningcited by 50
  6. 2021SmoothMix: Training Confidence-calibrated Smoothed Classifiers for Certified Robustnesscited by 49
  7. 2021Meta-Learning with Fewer Tasks through Task Interpolationcited by 45
  8. 20213D-VField: Adversarial Augmentation of Point Clouds for Domain Generalization in 3D Object Detectioncited by 43
  9. 2023Understanding Multimodal Contrastive Learning and Incorporating Unpaired Datacited by 43
  10. 2022Enhancing Adversarial Training with Second-Order Statistics of Weightscited by 42
  11. 2023Improved Probabilistic Image-Text Representationscited by 40
  12. 2021Towards Calibrated Model for Long-Tailed Visual Recognition from Prior Perspectivecited by 39
  13. 2024A Unified Replay-based Continuous Learning Framework for Spatio-Temporal Prediction on Streaming Datacited by 39
  14. 2021Adversarial Training Helps Transfer Learning via Better Representationscited by 36
  15. 2022UMIX: Improving Importance Weighting for Subpopulation Shift via Uncertainty-Aware Mixupcited by 34
  16. 2021DP-InstaHide: Provably Defusing Poisoning and Backdoor Attacks with Differentially Private Data Augmentationscited by 33
  17. 2021MixMo: Mixing Multiple Inputs for Multiple Outputs via Deep Subnetworkscited by 32
  18. 2022A Survey of Mix-based Data Augmentation: Taxonomy, Methods, Applications, and Explainabilitycited by 30
  19. 2022Geodesic Multi-Modal Mixup for Robust Fine-Tuningcited by 29
  20. 2022SelecMix: Debiased Learning by Contradicting-pair Samplingcited by 29
  21. 2023Randomized Adversarial Training via Taylor Expansioncited by 28
  22. 2021Noisy Feature Mixupcited by 27
  23. 2022A Unified Analysis of Mixed Sample Data Augmentation: A Loss Function Perspectivecited by 27
  24. 2021Rethinking Influence Functions of Neural Networks in the Over-parameterized Regimecited by 26
  25. 2021Survey: Image Mixing and Deleting for Data Augmentationcited by 25
  26. 2021On the Effect of Pruning on Adversarial Robustnesscited by 23
  27. 2022Robustness Implies Generalization via Data-Dependent Generalization Boundscited by 23
  28. 2022A Comprehensive Survey of Data Augmentation in Visual Reinforcement Learningcited by 23
  29. 2021Learning PAC-Bayes Priors for Probabilistic Neural Networkscited by 21
  30. 2021Towards Understanding the Data Dependency of Mixup-style Trainingcited by 19
  31. 2023The Benefits of Mixup for Feature Learningcited by 19
  32. 2023RankMixup: Ranking-Based Mixup Training for Network Calibrationcited by 19
  33. 2023Pseudo-Bag Mixup Augmentation for Multiple Instance Learning-Based Whole Slide Image Classificationcited by 18
  34. 2023MonoLSS: Learnable Sample Selection For Monocular 3D Detectioncited by 18
  35. 2024A Survey on Mixup Augmentations and Beyondcited by 18
  36. 2022NoisyMix: Boosting Model Robustness to Common Corruptionscited by 17
  37. 2023Toward Degree Bias in Embedding-Based Knowledge Graph Completioncited by 17
  38. 2023Last-Layer Fairness Fine-tuning is Simple and Effective for Neural Networkscited by 17

50 most cited of 193 All 193 (JSON)