Research map / Adversarial robustness and security / Certified and empirical robustness
FERD: Fairness-Enhanced Data-Free Adversarial Robustness Distillation
Zhengxiao Li, Liming Lu, Xu Zheng, Si Yuan Liang, Taric Chen, Yongbin Zhou, Shuchao Pang
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 robustness-guided class reweighting (ablation) | FERD · CIFAR-10 (WRN-34-10 teacher) ResNet-18 AutoAttack | worst-class robustness: 17.20 -> 20.80 | Improved |
| add robustness-guided class reweighting (ablation) | FERD · CIFAR-10 (WRN-34-10 teacher) ResNet-18 clean | worst-class accuracy: 63.20 -> 65.90 | Improved |
| add robustness-guided class reweighting (ablation) | FERD · CIFAR-10 (WRN-34-10 teacher) ResNet-18 clean | average accuracy: 79.53 -> 79.26 | Hurt |
| add FAEs (ablation) | FERD · CIFAR-10 (WRN-34-10 teacher) ResNet-18 AutoAttack | worst-class robustness: 19.90 -> 20.80 | Improved |
| add UTAEs (ablation) | FERD · CIFAR-10 (WRN-34-10 teacher) ResNet-18 AutoAttack | worst-class robustness: 20.00 -> 20.80 | Improved |
| add both FAEs and UTAEs (ablation) | FERD · CIFAR-10 (WRN-34-10 teacher) ResNet-18 AutoAttack | average robustness: 40.92 -> 42.24 | Improved |
| add both FAEs and UTAEs (ablation) | FERD · CIFAR-10 (WRN-34-10 teacher) ResNet-18 clean | worst-class accuracy: 62.50 -> 65.90 | Improved |
| use FERD reweighting instead of power-type reweighting (ablation) | FERD reweighting on CIFAR-10 ResNet-18 · CIFAR-10 (WRN-34-10 teacher) ResNet-18 PGD | worst-class robustness: 17.90 -> 20.40 | Improved |
| use FERD reweighting instead of linear reweighting (ablation) | FERD reweighting on CIFAR-10 ResNet-18 · CIFAR-10 (WRN-34-10 teacher) ResNet-18 FGSM | worst-class robustness: 29.10 -> 31.30 | Improved |
| use FERD reweighting instead of sigmoid or tanh reweighting (ablation) | FERD reweighting on CIFAR-10 ResNet-18 · CIFAR-10 (WRN-34-10 teacher) ResNet-18 clean | average accuracy: 80.16 -> 79.26 | Hurt |
| use penultimate residual module (block3) for non-robust feature distillation instead of block1 (ablation) | FERD · CIFAR-10 (WRN-34-10 teacher) ResNet-18 PGD-20 | worst-class robustness: 20.00 -> 23.50 | Improved |
| raise lambda_bn and lambda_uni from 1 to 5 (ablation) | FERD generator loss · CIFAR-10 (WRN-34-10 teacher) ResNet-18 AutoAttack | worst-class robustness: 16.90 -> 20.80 | Improved |
Closest papers by meaning
- Improving Adversarial Robust Fairness via Anti-Bias Soft Label Distillation (NeurIPS 2024 · similarity 0.91)
- On the Alignment between Fairness and Accuracy: from the Perspective of Adversarial Robustness (ICML 2025 · similarity 0.90)
- DAFA: Distance-Aware Fair Adversarial Training (ICLR 2024 · similarity 0.89)
- Out of Thin Air: Exploring Data-Free Adversarial Robustness Distillation (AAAI 2024 · similarity 0.89)
- Towards Fairness-Aware Adversarial Learning (CVPR 2024 · similarity 0.89)
- Towards Adversarially Robust Dataset Distillation by Curvature Regularization (AAAI 2025 · 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.
