Research map / Adversarial robustness and security / Certified and empirical robustness
On the Tradeoff Between Robustness and Fairness
Xinsong Ma, Zekai Wang, Weiwei Liu
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 |
|---|---|---|---|
| increase AT perturbation radius eps_train from 4/255 to 16/255 | Madry AT, ResNet18 · CIFAR-10, FGSM eps_test=16/255 | average robust accuracy: 29.04 -> 42.94 | Improved |
| increase AT perturbation radius eps_train from 4/255 to 16/255 | Madry AT, ResNet18 · CIFAR-10, FGSM eps_test=16/255 | variance of class-wise robust accuracy: 1.90 -> 3.52 | Hurt |
| increase AT perturbation radius eps_train from 4/255 to 16/255 | Madry AT, ResNet50 · CIFAR-10, FGSM eps_test=16/255 | variance of class-wise robust accuracy: 1.24 -> 3.89 | Hurt |
| increase AT perturbation radius eps_train from 4/255 to 16/255 | Madry AT, WRN-32-10 · CIFAR-10, PGD-20 eps_test=16/255 | average robust accuracy: 5.15 -> 17.05 | Improved |
| increase eps_train from 4/255 to 16/255 on the cat class | Madry AT, ResNet50 · CIFAR-10, FGSM eps_test=16/255 | class robust accuracy (cat): 12.0 -> 7.8 | Hurt |
| increase eps_train from 4/255 to 16/255 on the ship class | Madry AT, ResNet50 · CIFAR-10, FGSM eps_test=16/255 | class robust accuracy (ship): 34.2 -> 62.4 | Improved |
| increase eps_train from 4/255 to 16/255 on the deer class | Madry AT, ResNet50 · CIFAR-10, FGSM eps_test=16/255 | class robust accuracy (deer): 17.5 -> 13.7 | Hurt |
| add variance-of-class-wise-adversarial-risk regularizer (FAT) | Madry AT, ResNet18, eps_train=16/255 · CIFAR-10, FGSM eps_test=16/255 | variance of class-wise robust accuracy: 3.52 -> 3.32 | Improved |
| add FAT regularizer | Madry AT, ResNet18, eps_train=16/255 · CIFAR-10, PGD-20 eps_test=16/255 | variance of class-wise robust accuracy: 1.28 -> 1.06 | Improved |
| add FAT regularizer | Madry AT, ResNet18, eps_train=16/255 · CIFAR-10, C&W eps_test=16/255 | variance of class-wise robust accuracy: 1.38 -> 1.22 | Improved |
| add FAT regularizer | Madry AT, ResNet18, eps_train=12/255 · CIFAR-10, PGD-20 eps_test=20/255 | variance of class-wise robust accuracy: 0.35 -> 0.35 | No clear effect |
| add FAT regularizer | Madry AT, ResNet18, eps_train=16/255 · CIFAR-10, FGSM eps_test=16/255 | average robust accuracy: 42.97 -> 43.40 | No clear effect |
Closest papers by meaning
- On the Alignment between Fairness and Accuracy: from the Perspective of Adversarial Robustness (ICML 2025 · similarity 0.95)
- DAFA: Distance-Aware Fair Adversarial Training (ICLR 2024 · similarity 0.93)
- Towards Fairness-Aware Adversarial Learning (CVPR 2024 · similarity 0.91)
- Improving Adversarial Robust Fairness via Anti-Bias Soft Label Distillation (NeurIPS 2024 · similarity 0.91)
- On the Fairness ROAD: Robust Optimization for Adversarial Debiasing (ICLR 2024 · similarity 0.90)
- Can Private Machine Learning Be Fair? (AAAI 2025 · 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.
