Research map / Adversarial robustness and security / Certified and empirical robustness
Bilateral Adversarial Training: Towards Fast Training of More Robust Models Against Adversarial Attacks
Jianyu Wang, Haichao Zhang
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 random start to FGSM-based training (ablation) | one-step adversarial training · CIFAR10 CE7 attack | accuracy (%): 0.0 -> 48.0 | Improved |
| add random start to LL targeted training (ablation) | one-step adversarial training · CIFAR10 CE7 attack | accuracy (%): 0.0 -> 36.0 | Improved |
| use most-confusing target instead of least-likely target in training (ablation) | R-LL · CIFAR10 CE7 attack | accuracy (%): 36.0 -> 48.4 | Improved |
| add label smoothing (ablation) | R-FGSM · CIFAR10 CE20 | accuracy (%): 42.9 -> 49.0 | Improved |
| add label smoothing (ablation) | R-MC · CIFAR10 CE20 | accuracy (%): 43.4 -> 53.3 | Improved |
| replace label smoothing with adversarial label (ablation) | R-MC-LS · CIFAR10 CE20 | accuracy (%): 53.3 -> 55.3 | Improved |
| add label adversary to multi-step training (ablation) | Madry* · CIFAR10 CE20 | accuracy (%): 47.6 -> 53.2 | Improved |
| longer training (ablation) | R-MC-LA · CIFAR10 CE20 | accuracy (%): 55.3 -> 57.8 | Improved |
| train with perturbation budget 4 instead of 8 (ablation) | R-MC-LA · CIFAR10 CE100 | accuracy (%): 52.9 -> 60.1 | Improved |
| use weaker PGD2-8 attack in training instead of PGD7-2 (ablation) | Madry adversarial training · CIFAR10 PGD100-2 | accuracy (%): 47.2 -> 46.2 | No clear effect |
| increase training images per class from 0.5K to 5K (ablation) | R-MC-LA · CIFAR10 CE20 | accuracy (%): 25.1 -> 55.3 | Improved |
| use main-paper label solution instead of proportional-gradient solution (ablation) | R-MC-LA · CIFAR10 CE100 | accuracy (%): 51.1 -> 52.9 | Improved |
Closest papers by meaning
- Adversarial Training with Complementary Labels: On the Benefit of Gradually Informative Attacks (NeurIPS 2022 · similarity 0.90)
- Revisiting Adversarial Robustness Distillation: Robust Soft Labels Make Student Better (ICCV 2021 · similarity 0.89)
- Soften to Defend: Towards Adversarial Robustness via Self-Guided Label Refinement (CVPR 2024 · similarity 0.88)
- Adversarial Training Can Provably Improve Robustness: Theoretical Analysis of Feature Learning Process Under Structured Data (ICLR 2025 · similarity 0.88)
- Clarifying the Behavior and the Difficulty of Adversarial Training (AAAI 2024 · similarity 0.87)
- Enhancing the Adversarial Robustness via Manifold Projection (AAAI 2025 · 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.
