Research map / Adversarial robustness and security / Certified and empirical robustness
Adversarial training for free!
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Train with replay parameter m=8 rather than natural training (ablation) | Free adversarial training · CIFAR-10, PGD-20 | accuracy: 0.00% -> 46.82% | Helped |
| Increase replay from m=2 to m=8 (ablation) | Free adversarial training, Wide-ResNet 32-10 · CIFAR-10, PGD-20 | accuracy: 33.92% -> 46.82% | Helped |
| Increase replay from m=2 to m=8 (ablation) | Free adversarial training, Wide-ResNet 32-10 · CIFAR-10, natural images | accuracy: 91.45% -> 85.96% | Hurt |
| Increase replay from m=2 to m=8 (ablation) | Free adversarial training, Wide-ResNet 32-10 · CIFAR-100, PGD-20 | accuracy: 15.37% -> 25.88% | Helped |
| Increase replay from m=2 to m=8 (ablation) | Free adversarial training, Wide-ResNet 32-10 · CIFAR-100, natural images | accuracy: 69.20% -> 62.13% | Hurt |
| Increase replay from m=2 to m=4 (ablation) | Free adversarial training, ResNet-50 · ImageNet validation, PGD-50, ε=2 | accuracy: 36.340% -> 43.392% | Helped |
| Increase replay from m=2 to m=4 (ablation) | Free adversarial training, ResNet-50 · ImageNet validation, natural images, ε=2 | accuracy: 71.210% -> 64.446% | Hurt |
| Use ResNet-152 rather than ResNet-50 (ablation) | Free m=4 adversarial training · ImageNet validation, PGD-50, ε=4 | accuracy: 31.878% -> 36.044% | Helped |
| Use free training rather than 2-PGD training | ResNet-50 adversarial training · ImageNet validation, PGD-50, ε=4 | training time: 10,435 -> 3016 | Helped |
Closest papers by meaning
- Fast is better than free: Revisiting adversarial training (ICLR 2020 · similarity 0.92)
- On Adversarial Training without Perturbing all Examples (ICLR 2024 · similarity 0.91)
- Improving Adversarial Robustness Requires Revisiting Misclassified Examples (ICLR 2020 · similarity 0.90)
- Adversarial Training Can Provably Improve Robustness: Theoretical Analysis of Feature Learning Process Under Structured Data (ICLR 2025 · similarity 0.90)
- A Closer Look at Curriculum Adversarial Training: From an Online Perspective (AAAI 2024 · similarity 0.88)
- Taxonomy Driven Fast Adversarial Training (AAAI 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.