Research map / Adversarial robustness and security / Certified and empirical robustness
Label Noise in Adversarial Training: A Novel Perspective to Study Robust Overfitting
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 |
|---|---|---|---|
| self-distillation with automatically calibrated T and lambda | adversarial training · CIFAR-10 | robust acc at last checkpoint: 41.42 -> 48.80 | Helped |
| self-distillation with automatically calibrated T and lambda | adversarial training · CIFAR-100 | robust acc at last checkpoint: 19.75 -> 26.24 | Helped |
| self-distillation with automatically calibrated T and lambda | adversarial training · Tiny-ImageNet | robust acc at last checkpoint: 15.40 -> 18.39 | Helped |
| auto-determined T, lambda instead of fixed T=2, lambda=0.5 (ablation) | KD-AT · CIFAR-10 | robust overfitting gap: 2.43 -> 0.25 | Helped |
| auto-determined T, lambda instead of fixed T=2, lambda=0.5 (ablation) | KD-AT · CIFAR-10 | robust acc at best checkpoint: 48.76 -> 49.05 | Helped |
| auto-determined T, lambda instead of fixed T=2, lambda=0.5 (ablation) | KD-AT · CIFAR-100 | robust overfitting gap: 2.19 -> 0.12 | Helped |
| auto-determined T, lambda instead of fixed T=2, lambda=0.5 (ablation) | KD-AT · Tiny-ImageNet | robust acc at best checkpoint: 17.86 -> 18.29 | Helped |
| auto-determined T, lambda instead of fixed T=2, lambda=0.5 (ablation) | KD-AT · CIFAR-10 | standard acc at last checkpoint: 85.49 -> 84.47 | Hurt |
| increase perturbation radius in standard training on fixed adversarial set (ablation) | WRN-28-5 · CIFAR-10 | overfitting | Hurt |
Closest papers by meaning
- Soften to Defend: Towards Adversarial Robustness via Self-Guided Label Refinement (CVPR 2024 · similarity 0.93)
- A law of adversarial risk, interpolation, and label noise (ICLR 2023 · similarity 0.92)
- How Benign is Benign Overfitting ? (ICLR 2021 · similarity 0.91)
- Modeling Adversarial Noise for Adversarial Training (ICML 2022 · similarity 0.89)
- Annealing Self-Distillation Rectification Improves Adversarial Training (ICLR 2024 · similarity 0.88)
- On the Role of Label Noise in the Feature Learning Process (ICML 2025 · similarity 0.87)
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.