Research map / Adversarial robustness and security / Certified and empirical robustness
Prior-Guided Adversarial Initialization for Fast Adversarial Training
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 regularizer (ablation) | FGSM-BP · CIFAR-10 ResNet18 | AA accuracy: 43.30 -> 47.47 | Improved |
| add prior-guided regularizer (ablation) | FGSM-EP · CIFAR-10 ResNet18 | AA accuracy: 43.39 -> 47.94 | Improved |
| add prior-guided regularizer (ablation) | FGSM-MEP · CIFAR-10 ResNet18 | AA accuracy: 43.99 -> 49.00 | Improved |
| add regularizer to random-init FGSM (ablation) | FGSM-RS · CIFAR-10 ResNet18 | AA accuracy (best): 37.07 -> 46.80 | Improved |
| replace random initialization with previous-batch adversarial perturbation (ablation) | FGSM-RS · CIFAR-10 ResNet18 | PGD-50 accuracy (last checkpoint): 00.02 -> 44.39 | Improved |
| use momentum of all previous epochs instead of previous batch as initialization (ablation) | FGSM-BP w/o regularizer · CIFAR-10 ResNet18 | AA accuracy (best): 43.30 -> 43.99 | Improved |
| use momentum of all previous epochs instead of previous epoch as initialization (ablation) | FGSM-EP with regularizer · CIFAR-10 ResNet18 | AA accuracy (best): 47.94 -> 49.00 | Improved |
| use FGSM-MEP instead of random-init FGSM-RS | FGSM-RS · Tiny ImageNet last checkpoint | PGD-10 accuracy: 0.00 -> 22.02 | Improved |
| use FGSM-MEP instead of multi-step PGD-AT | PGD-AT · Tiny ImageNet best checkpoint | PGD-10 accuracy: 20.2 -> 23.8 | Improved |
| use FGSM-BP instead of FGSM-RS at epsilon 8 | FGSM-RS · ImageNet ResNet50 epsilon=8 | PGD-10 accuracy: 0.00 -> 21.76 | Improved |
| use FGSM-MEP instead of PGD-AT | PGD-AT · CIFAR-10 ResNet18 | AA accuracy: about 48% -> about 49% | Improved |
| use FGSM-MEP instead of NuAT for training time | NuAT · CIFAR-10 ResNet18 | training time (min): 104 -> 73 min | Improved |
Closest papers by meaning
- Improving Fast Adversarial Training With Prior-Guided Knowledge (TPAMI 2024 · similarity 0.95)
- Adversarial Initialization with Universal Adversarial Perturbation: A New Approach to Fast Adversarial Training (AAAI 2024 · similarity 0.94)
- Preventing Catastrophic Overfitting in Fast Adversarial Training: A Bi-level Optimization Perspective (ECCV 2024 · similarity 0.93)
- Taxonomy Driven Fast Adversarial Training (AAAI 2024 · similarity 0.88)
- Quadratic Upper Bound for Boosting Robustness (ICML 2025 · similarity 0.87)
- Rethinking Invariance Regularization in Adversarial Training to Improve Robustness-Accuracy Trade-off (ICLR 2025 · similarity 0.86)
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.
