Research map / Adversarial robustness and security / Black-box and transfer attacks
Efficient Adversarial Training With Transferable Adversarial Examples
Haizhong Zheng, Ziqi Zhang, Juncheng Gu, Honglak Lee, Atul Prakash
Check 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 |
|---|---|---|---|
| Reuse adversarial perturbations across epochs and use one attack iteration | MAT adversarial training · MNIST | PGD-40 accuracy: 96.21% -> 96.31% | No clear effect |
| Reuse adversarial perturbations across epochs and use one attack iteration | MAT adversarial training · MNIST | training time (sec): 3933 -> 297 | Improved |
| Reuse adversarial perturbations across epochs and use one attack iteration | MAT adversarial training · CIFAR10 | PGD-20 accuracy: 47.07% -> 50.96% | Improved |
| Reuse adversarial perturbations across epochs and use three attack iterations | TRADES adversarial training · CIFAR10 | PGD-20 accuracy: 56.6% -> 56.36% | No clear effect |
| Add inverse data augmentation when carrying perturbations across epochs (ablation) | MAT ATTA-1 · CIFAR10 | PGD-20 accuracy: 42.55% -> 50.96% | Improved |
| Add inverse data augmentation when carrying perturbations across epochs (ablation) | TRADES ATTA-1 · CIFAR10 | PGD-20 accuracy: 48.38% -> 54.50% | Improved |
| Remove data augmentation along with inverse data augmentation (ablation) | MAT ATTA-1 · CIFAR10 | natural accuracy: 85.71% -> 82.53% | Hurt |
| Use fixed-label MAT loss instead of TRADES loss for attack generation (ablation) | TRADES ATTA-1 · CIFAR10 | PGD-20 accuracy: 52.08% -> 54.50% | Improved |
Closest papers by meaning
- Evading Defenses to Transferable Adversarial Examples by Translation-Invariant Attacks (CVPR 2019 · similarity 0.90)
- On Adversarial Training without Perturbing all Examples (ICLR 2024 · similarity 0.90)
- Learning to Learn Transferable Attack (AAAI 2022 · similarity 0.90)
- Boosting Adversarial Transferability through Augmentation in Hypothesis Space (CVPR 2025 · similarity 0.89)
- Improving Transferability of Adversarial Examples With Input Diversity (CVPR 2019 · similarity 0.89)
- Enhancing Adversarial Transferability with Adversarial Weight Tuning (AAAI 2025 · 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.
