Research map / Adversarial robustness and security / Certified and empirical robustness
Improving Adversarial Robustness by Putting More Regularizations on Less Robust Samples
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 |
|---|---|---|---|
| Use ARoW's vulnerable-sample-weighted regularization instead of TRADES | adversarial training · CIFAR10, CIFAR100, SVHN, FMNIST | AutoAttack accuracy: CIFAR10: 54.31 -> 55.15; CIFAR100: 29.07 -> 30.42; SVHN: 51.06 -> 51.93; FMNIST: 88.19 -> 88.54 | Helped |
| Use ARoW instead of GAIR-AT | adversarial training · CIFAR10, SVHN | PGD^20 accuracy: CIFAR10: 67.27 -> 58.38; SVHN: 70.29 -> 61.14 | Hurt |
| Use ARoW instead of GAIR-AT | adversarial training · CIFAR10, SVHN | AutoAttack accuracy: CIFAR10: 46.41 -> 55.15; SVHN: 38.26 -> 51.93 | Helped |
| Use ARoW instead of HAT | CIFAR10 with extra data · CIFAR10, 80M-TI(500K), WRN-28-10 | AutoAttack accuracy: 60.96 -> 60.91 | No clear effect |
| Remove label smoothing (ablation) | ARoW · CIFAR10, WRN-34-10 | AutoAttack accuracy: 55.15 -> 55.01 | Hurt |
| Apply label smoothing (ablation) | TRADES · CIFAR10, WRN-34-10 | AutoAttack accuracy: 54.31 -> 54.66 | Helped |
| Apply SWA (ablation) | ARoW · CIFAR10, WRN-34-10 | AutoAttack accuracy: 52.95(0.06) -> 55.21(0.14) | Helped |
| Apply SWA (ablation) | MART · CIFAR10, WRN-34-10 | AutoAttack accuracy: 50.95(0.09) -> 48.94(0.09) | Hurt |
| Increase lambda from 2.5 to 5.0 (ablation) | ARoW · CIFAR10, ResNet-18 | AutoAttack accuracy: 49.66 -> 51.33 | Helped |
| Increase lambda from 2.5 to 5.0 (ablation) | ARoW · CIFAR10, ResNet-18 | standard accuracy: 85.30 -> 82.53 | Hurt |
| Set label-smoothing alpha to 0.40 rather than 0.05 (ablation) | ARoW · CIFAR10, ResNet-18 | AutoAttack accuracy: 49.88 -> 50.11 | Helped |
| Add AWP (ablation) | ARoW rather than TRADES · CIFAR10, ResNet-18 | AutoAttack accuracy: 49.56 -> 50.64 | Helped |
Closest papers by meaning
- Consistency Regularization for Adversarial Robustness (AAAI 2022 · similarity 0.93)
- Adversarial Robustness through Local Linearization (NeurIPS 2019 · similarity 0.92)
- Adversarial Training Can Provably Improve Robustness: Theoretical Analysis of Feature Learning Process Under Structured Data (ICLR 2025 · similarity 0.92)
- Towards Understanding the Regularization of Adversarial Robustness on Neural Networks (ICML 2020 · similarity 0.92)
- Attacks Which Do Not Kill Training Make Adversarial Learning Stronger (ICML 2020 · similarity 0.91)
- Regularization properties of adversarially-trained linear regression (NeurIPS 2023 · similarity 0.91)
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.