Research map / Label-efficient and robust learning / Semi-supervised learning
Enhancing Adversarial Robustness in Low-Label Regime via Adaptively Weighted Regularization and Knowledge Distillation
Dongyoon Yang, Insung Kong, Yongdai Kim
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 |
|---|---|---|---|
| use SRST-AWR instead of RST and UAT++ | semi-supervised adversarial training · CIFAR-10 (4,000 labels) | AutoAttack robust accuracy: 46.69 -> 51.72 (vs RST) | Improved |
| replace supervised teacher with semi-supervised teacher (ablation) | RST · CIFAR-10 (4,000 labels) | AutoAttack robust accuracy: 46.69 -> 49.32 | Improved |
| replace supervised teacher with semi-supervised teacher (ablation) | UAT++ · CIFAR-10 (4,000 labels) | AutoAttack robust accuracy: 46.53 -> 48.46 | Improved |
| add adaptive weight w (AWR) instead of constant weight (TRADES-style) (ablation) | SRST · CIFAR-10 (500 labels) | AutoAttack robust accuracy: 43.93 -> 50.76 | Improved |
| add adaptive weight w (AWR) instead of constant weight (ablation) | SRST · STL-10 (1,000 labels) | AutoAttack robust accuracy: 63.41 -> 72.51 | Improved |
| add knowledge distillation with soft pseudo labels (ablation) | SRST-AWR · CIFAR-10 (4,000 labels) | AutoAttack robust accuracy: 50.02 -> 51.72 | Improved |
| add knowledge distillation with soft pseudo labels (ablation) | SRST-AWR · CIFAR-10 (4,000 labels) | standard accuracy: 84.37 -> 84.56 | No clear effect |
| use SRST-AWR instead of ARMOURED | semi-supervised adversarial training · CIFAR-10 (WRN-28-2) | PGD-7 robust accuracy: 55.12 -> 52.43 | Hurt |
| use SRST-AWR instead of ARMOURED | semi-supervised adversarial training · SVHN (WRN-28-2) | AutoAttack robust accuracy: 23.35 -> 78.21 | Improved |
| train SRST-AWR with 12% of labels versus supervised TRADES with 100% labels | SRST-AWR · CIFAR-10 | AutoAttack robust accuracy: 51.72 -> 51.88 | Improved |
| use AWR instead of TRADES with fully labeled data | supervised adversarial training · CIFAR-10 (fully labeled) | AutoAttack robust accuracy: 51.72 -> 51.97 | Improved |
Closest papers by meaning
- Unlabeled Data Improves Adversarial Robustness (NeurIPS 2019 · similarity 0.91)
- Learning Better Certified Models from Empirically-Robust Teachers (NeurIPS 2026 · similarity 0.89)
- Robust Distillation via Untargeted and Targeted Intermediate Adversarial Samples (CVPR 2024 · similarity 0.89)
- Weakly Supervised Contrastive Adversarial Training for Learning Robust Features from Semi-supervised Data (CVPR 2025 · similarity 0.89)
- Soften to Defend: Towards Adversarial Robustness via Self-Guided Label Refinement (CVPR 2024 · similarity 0.88)
- Improving Adversarial Robust Fairness via Anti-Bias Soft Label Distillation (NeurIPS 2024 · similarity 0.88)
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.
