Research map / Adversarial robustness and security / Certified and empirical robustness
Adversarial Coreset Selection for Efficient Robust Training
Hadi M. Dolatabadi, Sarah Erfani, Christopher A. 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 |
|---|---|---|---|
| train on adversarial coreset (GradMatch) instead of full data | TRADES · CIFAR-10 ResNet-18 | robust accuracy (%): 44.19 -> 41.52 | Hurt |
| train on adversarial coreset (Craig) instead of full data | L_inf-PGD · CIFAR-10 ResNet-18 | robust accuracy (%): 41.39 -> 45.07 | Improved |
| train on adversarial coreset (Craig) instead of full data | L2-PGD · SVHN ResNet-18 | robust accuracy (%): 53.02 -> 49.68 | Hurt |
| train on adversarial coreset (GradMatch) instead of full data | TRADES · CIFAR-10 WideResNet-28-10 | robust accuracy (%): 45.08 -> 47.83 | Improved |
| train on adversarial coreset (GradMatch) instead of full data | L_inf-PGD · ImageNet-100 ResNet-34 | robust accuracy (%): 40.52 -> 40.72 | No clear effect |
| train on adversarial coreset (GradMatch) instead of full data | TRADES · CIFAR-10 ResNet-18 | training time (mins): 344.29 -> 178.73 | Improved |
| train on adversarial coreset (Craig) instead of full data | TRADES · CIFAR-10 ResNet-18 | clean accuracy (%): 85.41 -> 83.03 | Hurt |
| add adversarial coreset selection to perceptual adversarial training (GradMatch) | Fast-LPA PAT · ImageNet-12 | unseen robust accuracy (%): 53.23 -> 53.17 | No clear effect |
| add adversarial coreset selection (Craig) to Fast adversarial training | Fast AT · CIFAR-10 | training time (mins): 31.068 -> 16.695 | Improved |
| use only coreset samples rather than half coreset plus half clean (ablation) | adversarial coreset (GradMatch) L_inf-PGD · CIFAR-10 ResNet-18 | robust accuracy (%): 40.05 -> 45.23 | Improved |
| use only coreset samples rather than half coreset plus half clean (ablation) | adversarial coreset (GradMatch) L_inf-PGD · CIFAR-10 ResNet-18 | clean accuracy (%): 80.67 -> 84.31 | Hurt |
| use adversarial coreset selection instead of random data selection (ablation) | adversarial GradMatch TRADES · CIFAR-10 | relative robust error | Improved |
Closest papers by meaning
- l∞-Robustness and Beyond: Unleashing Efficient Adversarial Training (ECCV 2022 · similarity 0.93)
- Adversarial Training Can Provably Improve Robustness: Theoretical Analysis of Feature Learning Process Under Structured Data (ICLR 2025 · similarity 0.91)
- Ensemble Adversarial Training: Attacks and Defenses (ICLR 2018 · similarity 0.90)
- Analyzing the Implicit Bias of Adversarial Training From a Generalized Margin Perspective (TPAMI 2025 · similarity 0.90)
- Efficient Adversarial Contrastive Learning via Robustness-Aware Coreset Selection (NeurIPS 2023 · 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.
