Research map / Adversarial robustness and security / Certified and empirical robustness
Preventing Catastrophic Overfitting in Fast Adversarial Training: A Bi-level Optimization Perspective
Zhaoxin Wang, Handing Wang, Cong Tian, Yaochu Jin
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 adaptive fusion factor to loss + fusion (ablation) | FGSM-PCO · CIFAR10 (ResNet18) | PGD10 accuracy last checkpoint (%): 53.56 -> 56.12 | Improved |
| add regularization loss to fusion + adaptive (ablation) | FGSM-PCO · CIFAR10 (ResNet18) | PGD10 accuracy last checkpoint (%): 50.57 -> 56.12 | Improved |
| add perturbation fusion to loss + adaptive (ablation) | FGSM-PCO · CIFAR10 (ResNet18) | PGD10 accuracy last checkpoint (%): 48.81 -> 56.12 | Improved |
| use fusion alone without regularization loss (ablation) | FGSM-PCO · CIFAR10 (ResNet18) | PGD10 accuracy last checkpoint (%): 48.42 -> 10.18 | Hurt |
| FGSM-PCO vs FGSM-MEP under PGD10 on CIFAR10 | FGSM-MEP · CIFAR10 (ResNet18) | PGD10 accuracy (%): 55.13 -> 56.32 | Improved |
| FGSM-PCO vs FGSM-MEP under AutoAttack on CIFAR10 | FGSM-MEP · CIFAR10 (ResNet18) | AutoAttack accuracy (%): 48.23 -> 48.04 | No clear effect |
| FGSM-PCO vs FGSM-MEP on catastrophic overfitting frequency | FGSM-MEP · CIFAR100 (WideResNet34-10) | overfitting runs out of 10: 6/10 -> 0/10 | Improved |
| FGSM-PCO vs PGD-AT on CIFAR100 WRN34-10 | PGD-AT · CIFAR100 (WideResNet34-10) | PGD10 accuracy best checkpoint (%): 32.36 -> 29.80 | Hurt |
| FGSM-PCO vs FGSM-MEP at last checkpoint on Tiny-ImageNet | FGSM-MEP · Tiny-ImageNet (PreActResNet18) | PGD10 accuracy last checkpoint (%): 3.09 -> 14.55 | Improved |
| vary regularization coefficient beta between 5 and 15 (ablation) | FGSM-PCO · CIFAR10 (ResNet18) | accuracy | No clear effect |
Closest papers by meaning
- Improving Fast Adversarial Training With Prior-Guided Knowledge (TPAMI 2024 · similarity 0.93)
- Prior-Guided Adversarial Initialization for Fast Adversarial Training (ECCV 2022 · similarity 0.93)
- Adversarial Initialization with Universal Adversarial Perturbation: A New Approach to Fast Adversarial Training (AAAI 2024 · similarity 0.90)
- Taxonomy Driven Fast Adversarial Training (AAAI 2024 · similarity 0.90)
- Quadratic Upper Bound for Boosting Robustness (ICML 2025 · similarity 0.88)
- Efficient local linearity regularization to overcome catastrophic overfitting (ICLR 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.
