Research map / Label-efficient and robust learning / Test-time adaptation
AugMix: A Simple Data Processing Method to Improve Robustness and Uncertainty
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| replace standard training with AugMix | WideResNet · CIFAR-10-C | error rate: 26.9 -> 11.2 | Helped |
| replace standard training with AugMix | WideResNet · CIFAR-100-C | error rate: 53.3 -> 35.9 | Helped |
| replace standard training with AugMix | ResNet-50 · ImageNet-C | mCE: 80.6 -> 68.4 | Helped |
| replace standard training with AugMix | ResNet-50 · ImageNet-P | mFR: 57.2 -> 37.4 | Helped |
| use randomly sampled augmentation chains (ablation) | WideResNet · CIFAR-10-C | error rate: 26.9 -> 17.0 | Helped |
| add JSD consistency loss to random augmentation chains (ablation) | WideResNet · CIFAR-10-C | error rate: 17.0 -> 14.7 | Helped |
| add augmentation mixing without JSD loss (ablation) | WideResNet · CIFAR-10-C | error rate: 26.9 -> 13.1 | Helped |
| add JSD consistency loss to AugmentAndMix (ablation) | WideResNet · CIFAR-10-C | error rate: 13.1 -> 11.2 | Helped |
| add more mixing by applying AugMix on top of Mixup (ablation) | AugMix · CIFAR-10-C | error rate: 11.2 -> 13.3 | Hurt |
| combine AugMix with Stylized ImageNet training | ResNet-50 · ImageNet-C | mCE: 68.4 -> 64.9 | Helped |
| combine AugMix with Stylized ImageNet training | ResNet-50 · ImageNet-P | mFR: 37.4 -> 38.9 | Hurt |
Closest papers by meaning
- IPMix: Label-Preserving Data Augmentation Method for Training Robust Classifiers (NeurIPS 2023 · similarity 0.91)
- SURE: SUrvey REcipes for building reliable and robust deep networks (CVPR 2024 · similarity 0.88)
- DiffAug: A Diffuse-and-Denoise Augmentation for Training Robust Classifiers (NeurIPS 2024 · similarity 0.86)
- Improving robustness to corruptions with multiplicative weight perturbations (NeurIPS 2024 · similarity 0.85)
- Adversarial AutoMixup (ICLR 2024 · similarity 0.84)
- Foundation Model-oriented Robustness: Robust Image Model Evaluation with Pretrained Models (ICLR 2024 · similarity 0.84)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.