Research map / Adversarial robustness and security / Certified and empirical robustness
l∞-Robustness and Beyond: Unleashing Efficient Adversarial Training
Hadi M. Dolatabadi, Sarah Erfani, Christopher Leckie
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 |
|---|---|---|---|
| adversarial coreset selection (Craig, 50% data) instead of full training | TRADES · CIFAR-10 ResNet-18 | training time (min): 344.29 -> 179.20 | Improved |
| adversarial coreset selection (Craig) instead of full training | TRADES · CIFAR-10 ResNet-18 | robust accuracy (%): 44.19 -> 41.45 | Hurt |
| adversarial coreset selection (GradMatch) instead of full training | l-inf PGD · CIFAR-10 ResNet-18 | robust accuracy (%): 41.39 -> 45.23 | Improved |
| adversarial coreset selection (GradMatch) instead of full training | l-inf PGD · CIFAR-10 ResNet-18 | clean accuracy (%): 83.14 -> 80.67 | Hurt |
| adversarial coreset selection (GradMatch, 30% data) instead of full training | l2-PGD · SVHN ResNet-18 | training time (min): 389.46 -> 125.53 | Improved |
| adversarial coreset selection (GradMatch) instead of full training | l2-PGD · SVHN ResNet-18 | robust accuracy (%): 53.02 -> 50.41 | Hurt |
| adversarial coreset selection (Craig) instead of full PAT | Fast-LPA · CIFAR-10 ResNet-50 | training time (min): 1682.94 -> 767.34 | Improved |
| adversarial coreset selection (GradMatch) instead of full PAT | Fast-LPA · ImageNet-12 ResNet-50 | unseen robust accuracy (%): 53.23 -> 53.17 | No clear effect |
| add adversarial coreset selection (GradMatch) to Fast Adversarial Training | FAT · CIFAR-10 ResNet-18 | training time (min): 31.068 -> 16.419 | Improved |
| add adversarial coreset selection (GradMatch) to FAT | FAT · CIFAR-10 ResNet-18 | clean accuracy (%): 86.20 -> 82.53 | Hurt |
| add adversarial coreset selection to FAT | FAT · CIFAR-10 ResNet-18 | robust accuracy (%): 47.54 -> 47.88 | No clear effect |
| select coreset by gradient matching instead of random selection (ablation) | adversarial GradMatch · CIFAR-10 TRADES | relative robust error | Improved |
Closest papers by meaning
- Adversarial Coreset Selection for Efficient Robust Training (IJCV 2023 · similarity 0.93)
- Adversarial Training Can Provably Improve Robustness: Theoretical Analysis of Feature Learning Process Under Structured Data (ICLR 2025 · similarity 0.91)
- Adversarial Training and Robustness for Multiple Perturbations (NeurIPS 2019 · similarity 0.91)
- Robust Models are less Over-Confident (NeurIPS 2022 · similarity 0.91)
- Quadratic Upper Bound for Boosting Robustness (ICML 2025 · similarity 0.90)
- Learning Better Certified Models from Empirically-Robust Teachers (NeurIPS 2026 · 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.
