Research map / Adversarial robustness and security / Certified and empirical robustness
Exploring and Exploiting Decision Boundary Dynamics for Adversarial Robustness
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 |
|---|---|---|---|
| Dynamics-Aware Robust Training (DyART) replacing standard adversarial training objective | WRN-28-10 on CIFAR-10 · CIFAR-10 AutoAttack | robust accuracy at eps=8/255: 71.73 vs 71.24 (AT) | Helped |
| DyART objective instead of baselines | ResNet-18 on Tiny-ImageNet · Tiny-ImageNet AutoAttack | clean and robust accuracy: clean 49.71 vs 48.09 (AT); robust eps=8/255 18.02 vs 17.79 (MART) | Helped |
| DyART objective instead of TRADES at large perturbation bound | WRN-28-10 on CIFAR-10 · CIFAR-10 AutoAttack | robust accuracy at eps=16/255: 20.79 vs 21.59 (TRADES) | Hurt |
| DyART objective instead of GAIRAT/AT on clean accuracy | WRN-28-10 on CIFAR-10 · CIFAR-10 AutoAttack | clean accuracy: 85.55 vs 86.59 (GAIRAT), 85.65 (AT) | Hurt |
| Prioritizing smaller margins via convex decreasing cost h(R) with soft margins (ablation) | DyART training dynamics · Naturally pretrained model, margins below 4/255 | proportion of negative-speed points: 29.2% to 15.3% | Helped |
| Adversarial training's decision boundary dynamics (conflicting movement) observed | AT on ResNet-18 (GN) CIFAR-10 · training batch margins below 8/255 | proportion of negative-speed points: 28.8% and 29.4% | No clear effect |
Closest papers by meaning
- Hold me tight! Influence of discriminative features on deep network boundaries (NeurIPS 2020 · similarity 0.91)
- Detecting Brittle Decisions for Free: Leveraging Margin Consistency in Deep Robust Classifiers (NeurIPS 2024 · similarity 0.88)
- Out of Thin Air: Exploring Data-Free Adversarial Robustness Distillation (AAAI 2024 · similarity 0.86)
- Layer-Aware Analysis of Catastrophic Overfitting: Revealing the Pseudo-Robust Shortcut Dependency (ICML 2024 · similarity 0.86)
- Robust Distillation via Untargeted and Targeted Intermediate Adversarial Samples (CVPR 2024 · similarity 0.86)
- Adversarial Training Can Provably Improve Robustness: Theoretical Analysis of Feature Learning Process Under Structured Data (ICLR 2025 · similarity 0.86)
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.