Research map / Adversarial robustness and security / Certified and empirical robustness
Data augmentation alone can improve adversarial training
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 |
|---|---|---|---|
| Use IDBH instead of baseline augmentation | PRN18 adversarial training · CIFAR10 | end robustness (AA, %): 42.46 -> 49.99 | Helped |
| Use IDBH instead of baseline augmentation | PRN18 adversarial training · SVHN | end robustness (AA, %): +13.23% | Helped |
| Use IDBH instead of baseline augmentation | PRN18 adversarial training · TIN | end robustness (AutoPGD, %): 13.82 -> 19.22 | Helped |
| Add SWA to IDBH | PRN18 adversarial training · CIFAR10 | end accuracy (%): 85.82 -> 86.45 | Helped |
| Add AWP and SWA to IDBH[weak] | PRN18 adversarial training · CIFAR10 | best robustness (AA, %): 50.34 -> 52.46 | Helped |
| Replace Padcrop with Cropshift (ablation) | IDBH · CIFAR10, PRN18 | end robustness (AA, %): 49.25 -> 49.99 | Helped |
| Replace Padcrop with Cropshift (ablation) | IDBH · CIFAR10, PRN18 + SWA | end robustness (AA, %): 49.40 -> 49.88 | Helped |
| Increase augmentation type diversity (ablation) | Adversarial training · CIFAR10 | end robustness | Helped |
| Increase spatial diversity of Cutout-i or Cropshift (ablation) | Adversarial training · CIFAR10 | end robustness | Helped |
| Increase strength diversity for Cutout-1, Cropshift-1, or Rotate (ablation) | Adversarial training · CIFAR10 | robust overfitting | Helped |
| Increase strength diversity for transformations other than Cutout-1, Cropshift-1, and Rotate (ablation) | Adversarial training · CIFAR10 | robustness | No clear effect |
| Increase augmentation hardness beyond the model's useful range (ablation) | Adversarial training · CIFAR10 | robustness | Hurt |
Closest papers by meaning
- Data Augmentation Can Improve Robustness (NeurIPS 2021 · similarity 0.93)
- Consistency Regularization for Adversarial Robustness (AAAI 2022 · similarity 0.91)
- Efficient and Effective Augmentation Strategy for Adversarial Training (NeurIPS 2022 · similarity 0.91)
- AugMax: Adversarial Composition of Random Augmentations for Robust Training (NeurIPS 2021 · similarity 0.90)
- Understanding Robust Overfitting of Adversarial Training and Beyond (ICML 2022 · similarity 0.89)
- Revisiting Adversarial Training Under Long-Tailed Distributions (CVPR 2024 · similarity 0.89)
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.