Research map / Adversarial robustness and security / Certified and empirical robustness
Adversarial Training for Probabilistic Robustness
Yi Zhang, Yuhang Chen, Zhen Chen, Wenjie Ruan, Xiaowei Huang, Siddartha Khastgir, Xingyu Zhao
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 |
|---|---|---|---|
| train with AT-PR instead of AT-WCR (PGD) | ResNet-18 · CIFAR-100 PR(eps=0.15) | PR: 80.47 -> 87.38 | Improved |
| train with AT-PR instead of AT-WCR (PGD) | ResNet-18 · CIFAR-10 PR(eps=0.15) | PR: 89.78 -> 93.21 | Improved |
| train with AT-PR instead of AT-WCR (PGD) | ResNet-18 · SVHN PR(eps=0.15) | PR: 95.41 -> 97.84 | Improved |
| train with AT-PR instead of AT-WCR (PGD) | WRN-50-2 · SVHN PR(eps=0.15) | PR: 96.77 -> 96.26 | Hurt |
| train with AT-PR instead of AT-WCR (PGD) | ResNet-18 · CIFAR-10 PGD accuracy | PGD accuracy: 45.01 -> 39.77 | Hurt |
| train with AT-PR instead of AT-WCR (PGD) | ResNet-34 · Tiny-ImageNet clean accuracy | accuracy: 49.33 -> 53.98 | Improved |
| replace PGD AE with farthest-from-boundary AE (AT-PR) vs random AE from Step 1 (ablation) | AT-PR · CIFAR-10 PR(eps=0.15) | PR: 92.64 -> 93.21 | Improved |
| replace PGD AE with farthest-from-boundary AE (AT-PR) vs random AE from Step 1 (ablation) | AT-PR · CIFAR-100 PR(eps=0.15) | PR: 85.21 -> 87.38 | Improved |
| use AT-PR instead of PGD for lower train-test PR gap | ResNet-18 · CIFAR-100 PR gap (eps=0.15) | generalization gap: 8.54 -> 2.87 | Improved |
| use AT-PR instead of PGD for lower train-test PR gap | ResNet-18 · Tiny-ImageNet PR gap (eps=0.15) | generalization gap: 11.94 -> 7.66 | Improved |
| add AT-PR fine-tuning to TRADES | ViT-B · CIFAR-10 | clean accuracy: 83.55 -> 86.03 | Improved |
| add AT-PR fine-tuning to TRADES | ViT-B · CIFAR-10 | PGD-20 accuracy: 53.14 -> 52.94 | Hurt |
Closest papers by meaning
- Probabilistically Robust Learning: Balancing Average and Worst-case Performance (ICML 2022 · similarity 0.91)
- Adversarial Robustness with Semi-Infinite Constrained Learning (NeurIPS 2021 · similarity 0.91)
- WAT: Improve the Worst-Class Robustness in Adversarial Training (AAAI 2023 · similarity 0.91)
- Bag of Tricks for Adversarial Training (ICLR 2021 · similarity 0.90)
- Adversarial Training Can Provably Improve Robustness: Theoretical Analysis of Feature Learning Process Under Structured Data (ICLR 2025 · similarity 0.90)
- On the Duality Between Sharpness-Aware Minimization and Adversarial Training (ICML 2024 · similarity 0.90)
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.
