Research map / Adversarial robustness and security / Certified and empirical robustness
One Size Does NOT Fit All: Data-Adaptive Adversarial Training
Shuo Yang, Chang Xu
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (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 |
|---|---|---|---|
| add data-adaptive perturbation size (DAAT) (ablation) | standard AT · CIFAR10 | clean accuracy (%): 84.66% -> 88.31% | Improved |
| add DAAT (ablation) | standard AT · CIFAR10 | PGD-20 robust accuracy (%): 47.83% -> 48.89% | Improved |
| add DAAT to TRADES (ablation) | TRADES · CIFAR10 | clean accuracy (%): 81.98% -> 83.55% | Improved |
| add DAAT to MART (ablation) | MART · CIFAR10 | clean accuracy (%): 80.67% -> 83.87% | Improved |
| add DAAT to MART (ablation) | MART · CIFAR10 | AutoAttack robust accuracy (%): 49.73% -> 49.82% | No clear effect |
| add DAAT to TRADES (ablation) | TRADES · SVHN | clean accuracy (%): 88.99% -> 90.28% | Improved |
| add DAAT to MART (ablation) | MART · SVHN | clean accuracy (%): 89.91% -> 92.16% | Improved |
| add DAAT to TRADES (ablation) | TRADES · SVHN | AutoAttack robust accuracy (%): 50.82% -> 50.59% | No clear effect |
| use DAAT instead of FAT | FAT · CIFAR10 | PGD-20 robust accuracy (%): 46.52% -> 48.89% | Improved |
| enlarge adaptive perturbation upper bound to 1.25 eps (ablation) | DAAT · CIFAR10 | PGD-20 robust accuracy (%): 48.89% -> 51.72%; clean 88.31% -> 86.93% | Mixed |
| use dynamic margin rho instead of fixed rho (ablation) | DAAT · CIFAR10 | natural and FGSM accuracy | Improved |
| increase margin rho (ablation) | DAAT · CIFAR10 | natural vs robust accuracy | Mixed |
Closest papers by meaning
- Strength-Adaptive Adversarial Training (TPAMI 2026 · similarity 0.91)
- Attacks Which Do Not Kill Training Make Adversarial Learning Stronger (ICML 2020 · similarity 0.89)
- AROID: Improving Adversarial Robustness Through Online Instance-Wise Data Augmentation (IJCV 2024 · similarity 0.88)
- Can Adversarial Training Be Manipulated By Non-Robust Features? (NeurIPS 2022 · similarity 0.88)
- Adversarial Training Can Provably Improve Robustness: Theoretical Analysis of Feature Learning Process Under Structured Data (ICLR 2025 · similarity 0.88)
- Taxonomy Driven Fast Adversarial Training (AAAI 2024 · similarity 0.88)
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.
