Research map / Adversarial robustness and security / Certified and empirical robustness
WAT: Improve the Worst-Class Robustness in 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 WAT instead of CSL | ResNet-18 · CIFAR-10, PGD-100 | worst-class robust accuracy: 33.8 -> 36.6 | Helped |
| Use WAT instead of CSL | ResNet-18 · CIFAR-10, CW | worst-class robust accuracy: 31.3 -> 33.3 | Helped |
| Use WAT instead of CSL | ResNet-18 · CIFAR-10, AutoAttack | worst-class robust accuracy: 29.2 -> 30.1 | Helped |
| Use WAT instead of CSL | ResNet-18 · CIFAR-100, PGD-100 | worst-class robust accuracy: 4.00 -> 5.00 | Helped |
| Use WAT instead of CSL | ResNet-18 · CIFAR-100, CW | worst-class robust accuracy: 2.00 -> 3.00 | Helped |
| Use WAT instead of CSL | ResNet-18 · CIFAR-100, AutoAttack | worst-class robust accuracy: 2.00 -> 3.00 | Helped |
| Use WAT instead of the strongest listed baseline | WideResNet-34-10 · CIFAR-10, PGD-100 | worst-class robust accuracy: 32.9 -> 34.9 | Helped |
| Use WAT instead of the strongest listed baseline | WideResNet-34-10 · CIFAR-10, CW | worst-class robust accuracy: 32.2 -> 33.4 | Helped |
| Use WAT instead of CSL | WideResNet-34-10 · CIFAR-10, AutoAttack | worst-class robust accuracy: 29.4 -> 30.7 | Helped |
| Set η to 0.01 rather than 0.1 (ablation) | WAT, ResNet-18 · CIFAR-10, PGD-100 | worst-class robust accuracy: 36.6 -> 26.6 | Hurt |
| Set η to 0.05 rather than 0.1 (ablation) | WAT, ResNet-18 · CIFAR-10, PGD-100 | worst-class robust accuracy: 36.6 -> 34.2 | Hurt |
| Set η to 0.5 rather than 0.1 (ablation) | WAT, ResNet-18 · CIFAR-10, PGD-100 | worst-class robust accuracy: 36.6 -> 37.5 | Helped |
Closest papers by meaning
- Improving Adversarial Robustness Requires Revisiting Misclassified Examples (ICLR 2020 · similarity 0.92)
- Robust Models are less Over-Confident (NeurIPS 2022 · similarity 0.92)
- Adversarial Training Can Provably Improve Robustness: Theoretical Analysis of Feature Learning Process Under Structured Data (ICLR 2025 · similarity 0.91)
- A Simple Class-Agnostic Approach to Enhance Fair Adversarial Training (NeurIPS 2026 · similarity 0.91)
- Identifying and Understanding Cross-Class Features in Adversarial Training (ICML 2025 · similarity 0.90)
- A Closer Look at Curriculum Adversarial Training: From an Online Perspective (AAAI 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.