Research map / Adversarial robustness and security / Certified and empirical robustness
Improving Fast Adversarial Training With Prior-Guided Knowledge
Xiaojun Jia, Yong Zhang, Xingxing Wei, Baoyuan Wu, Ke Ma, Jue Wang, Xiaochun Cao
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 prior-guided initialization (ablation) | FGSM-RS · CIFAR-10 ResNet18 | AutoAttack accuracy: 0.00 -> 44.92 | Improved |
| add prior-guided regularizer (ablation) | FGSM-PGK initialization-only · CIFAR-10 ResNet18 | AutoAttack accuracy: 44.92 -> 49.02 | Improved |
| add prior-guided regularizer (ablation) | FGSM-PGK initialization-only · CIFAR-10 ResNet18 | training time (h): 0.9 -> 1.2 | Hurt |
| add prior-guided regularizer (ablation) | FGSM-PGK initialization-only · CIFAR-10 ResNet18 | clean accuracy: 84.92 -> 81.77 | Hurt |
| add prior-guided ensemble weights (ablation) | FGSM-PGK initialization-only · CIFAR-10 ResNet18 | AutoAttack accuracy: 44.92 -> 46.08 | Improved |
| add prior-guided weights to initialization plus regularizer (ablation) | FGSM-PGK · CIFAR-10 ResNet18 | AutoAttack accuracy: 49.02 -> 49.51 | Improved |
| replace fixed-decay EMA with dynamic decay rate EMA (D-EMA) (ablation) | FGSM-RS-EMA · CIFAR-10 ResNet18 | AutoAttack accuracy (best): 45.15 -> 46.63 | Improved |
| replace fixed-decay EMA with D-EMA (ablation) | FGSM-RS-EMA · CIFAR-10 ResNet18 | AutoAttack accuracy (last): 0.0 -> 46.57 | Improved |
| use random initialization instead of zero initialization | FGSM-AT · CIFAR-10 | catastrophic overfitting epoch: 20-th epoch -> 70-th epoch | Mixed |
| use FGSM-PGK instead of multi-step PGD-AT | ResNet18 · CIFAR-10 | AutoAttack accuracy: 48.93 -> 49.51 | Improved |
| use FGSM-PGK instead of multi-step PGD-AT | ResNet18 · CIFAR-10 | training time (h): 4.4 -> 1.2 | Improved |
| use FGSM-PGK instead of FGSM-PGI | WideResNet34-10 · CIFAR-10 | PGD-50 accuracy: 56.4 -> 59.79 | Improved |
Closest papers by meaning
- Prior-Guided Adversarial Initialization for Fast Adversarial Training (ECCV 2022 · similarity 0.95)
- Preventing Catastrophic Overfitting in Fast Adversarial Training: A Bi-level Optimization Perspective (ECCV 2024 · similarity 0.93)
- Adversarial Initialization with Universal Adversarial Perturbation: A New Approach to Fast Adversarial Training (AAAI 2024 · similarity 0.91)
- Taxonomy Driven Fast Adversarial Training (AAAI 2024 · similarity 0.89)
- Quadratic Upper Bound for Boosting Robustness (ICML 2025 · similarity 0.88)
- Efficient local linearity regularization to overcome catastrophic overfitting (ICLR 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.
