Research map / Adversarial robustness and security / Certified and empirical robustness
Improving Adversarial Robustness by Mitigating Instability through Relearning
Yi Zeng, Ling Zhou, Ruilong Yu, Qihe Liu, Shijie Zhou
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 relearning (RE) to PGD-AT in place of TE | PGD-AT · CIFAR-10 ResNet-18 PGD-20 final | robust accuracy: 53.87 -> 55.40 | Improved |
| add relearning to PGD-AT | PGD-AT · CIFAR-10 ResNet-18 Auto Attack final | robust accuracy: 40.74 -> 50.83 | Improved |
| add relearning to PGD-AT | PGD-AT · CIFAR-10 ResNet-18 | RSR: -88.61 -> -22.80 | Improved |
| add relearning to PGD-AT on ViT | PGD-AT · CIFAR-10 ViT PGD-20 final | robust accuracy: 38.78 -> 44.55 | Improved |
| add relearning to MART | MART · CIFAR-10 ResNet-18 Auto Attack | robust accuracy: 43.61 -> 49.30 | Improved |
| add relearning to TRADES | TRADES · CIFAR-10 ResNet-18 Auto Attack | robust accuracy: 47.29 -> 49.76 | Improved |
| add relearning to AT-AWP | AT-AWP · CIFAR-100 ResNet-18 Auto Attack | robust accuracy: 25.30 -> 25.87 | Improved |
| add relearning to PGD-AT | PGD-AT · CIFAR-100 ResNet-18 Auto Attack | robust accuracy: 19.69 -> 26.00 | Improved |
| use relearning loss in adversarial example generation only (ablation) | PGD-AT · CIFAR-10 ResNet-18 PGD-20 final | robust accuracy: 41.40 -> 51.58 | Improved |
| use relearning loss in training only (ablation) | PGD-AT · CIFAR-10 ResNet-18 PGD-20 final | robust accuracy: 41.40 -> 49.08 | Improved |
| use relearning loss in both generation and training instead of one phase (ablation) | PGD-AT+RE · CIFAR-10 ResNet-18 PGD-20 final | robust accuracy: 51.58 -> 55.40 | Improved |
| increase relearning weight omega beyond 30 (ablation) | PGD-AT+RE · CIFAR-10 ResNet-18 | robust accuracy | Hurt |
Closest papers by meaning
- Balance, Imbalance, and Rebalance: Understanding Robust Overfitting from a Minimax Game Perspective (NeurIPS 2023 · similarity 0.92)
- Can Adversarial Training Be Manipulated By Non-Robust Features? (NeurIPS 2022 · similarity 0.91)
- Adversarial Training Can Provably Improve Robustness: Theoretical Analysis of Feature Learning Process Under Structured Data (ICLR 2025 · similarity 0.91)
- Conserve-Update-Revise to Cure Generalization and Robustness Trade-off in Adversarial Training (ICLR 2024 · similarity 0.90)
- Annealing Self-Distillation Rectification Improves Adversarial Training (ICLR 2024 · similarity 0.89)
- Adversarial Training Should Be Cast as a Non-Zero-Sum Game (ICLR 2024 · similarity 0.89)
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.
