Research map / Adversarial robustness and security / Certified and empirical robustness
Mitigating Error Amplification in Fast Adversarial Training
Mengnan Zhao, Lihe Zhang, Bo Wang, Tianhang Zheng, Hong Zhong, Geyong Min
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 perturbation budget allocation (PBA) (ablation) | DDG · CIFAR-10 (best) | clean accuracy: 81.70 -> 82.67 | Improved |
| add perturbation budget allocation (PBA) (ablation) | DDG · CIFAR-10 (best) | APGD accuracy: 57.46 -> 59.48 | Improved |
| add supervision signal adjustment (SSA) (ablation) | DDG · CIFAR-10 (best) | clean accuracy: 80.38 -> 82.67 | Improved |
| add supervision signal adjustment (SSA) (ablation) | DDG · CIFAR-10 (best) | C&W20 accuracy: 50.87 -> 49.86 | Hurt |
| add gradient smoothness regularization (GS) (ablation) | DDG · CIFAR-10 (best) | PGD-10 accuracy: 58.71 -> 60.44 | Improved |
| add gradient smoothness regularization (GS) (ablation) | DDG · CIFAR-10 (best) | clean accuracy: 84.32 -> 82.67 | Hurt |
| apply DDG instead of TDAT | ResNet-18 FAT · CIFAR-10 (best) | PGD-50 accuracy: 55.41 -> 59.33 | Improved |
| apply DDG instead of TDAT | ResNet-18 FAT · CIFAR-100 (best) | PGD-10 accuracy: 33.56 -> 34.42 | Improved |
| apply DDG instead of TDAT | ResNet-18 FAT · Tiny-ImageNet (best) | PGD-10 accuracy: 23.98 -> 24.35 | Improved |
| apply DDG instead of TDAT | ResNet-18 FAT · CIFAR-10 (best) | AutoAttack accuracy: 48.06 -> 48.08 | No clear effect |
| scale negative suppression term from 0 to 1.0 (ablation) | y_sr · CIFAR-10 | clean accuracy: 81.50 -> 82.94 | Improved |
| scale negative suppression term from 0 to 1.0 (ablation) | y_sr · CIFAR-10 | C&W10 accuracy: 51.02 -> 50.75 | Hurt |
Closest papers by meaning
- Improving Fast Adversarial Training With Prior-Guided Knowledge (TPAMI 2024 · similarity 0.90)
- Catastrophic Overfitting: A Potential Blessing in Disguise (ECCV 2024 · similarity 0.89)
- Layer-Aware Analysis of Catastrophic Overfitting: Revealing the Pseudo-Robust Shortcut Dependency (ICML 2024 · similarity 0.89)
- Improving Adversarial Robustness by Mitigating Instability through Relearning (ECCV 2026 · similarity 0.89)
- Preventing Catastrophic Overfitting in Fast Adversarial Training: A Bi-level Optimization Perspective (ECCV 2024 · similarity 0.88)
- Taxonomy Driven Fast Adversarial Training (AAAI 2024 · similarity 0.87)
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.
