Research map / Adversarial robustness and security / Certified and empirical robustness
Boosting Accuracy and Robustness of Student Models via Adaptive Adversarial Distillation
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 |
|---|---|---|---|
| Use adaptive inner optimization to maximize student-teacher prediction discrepancy, then distill at the resulting point | ResNet-18 with WRN-34-20 teacher · CIFAR-10 | AA accuracy (%): 49.85 -> 52.96 | Helped |
| Use adaptive inner optimization to maximize student-teacher prediction discrepancy, then distill at the resulting point | ResNet-18 with WRN-34-10 teacher · CIFAR-10 | Clean accuracy (%): 84.69 -> 86.75 | Helped |
| Use adaptive inner optimization to maximize student-teacher prediction discrepancy, then distill at the resulting point | ResNet-18 with WRN-34-10 teacher · CIFAR-10 | AA accuracy (%): 48.82 -> 50.74 | Helped |
| Combine adaptive adversarial distillation with IAD | ResNet-18 · CIFAR-100 | AA accuracy (%): 27.77 -> 27.98 | Helped |
| Use AdaAD instead of the strongest reported baseline | MobileNetV2 · CIFAR-100 | Clean accuracy (%): 57.18 -> 61.44 | Helped |
| Combine adaptive adversarial distillation with IAD | MobileNetV2 · CIFAR-100 | AA accuracy (%): 24.65 -> 26.43 | Helped |
| Use adaptive adversarial distillation with transfer-based attacks | ResNet-18 · CIFAR-10, attack transferred from ResNet-34 | PGD accuracy (%): 63.20 -> 64.57 | Helped |
| Increase the inner-optimization perturbation limit from 8/255 to 32/255 (ablation) | AdaAD ResNet-18 with WRN-34-20 teacher · CIFAR-10 | AA accuracy (%): 50.06% to 52.86% | Helped |
| Increase the inner-optimization perturbation limit from 8/255 to 32/255 (ablation) | AdaAD ResNet-18 · CIFAR-10 | Clean accuracy (%) | No clear effect |
| Increase the inner-optimization perturbation limit (ablation) | PGD-AT, TRADES, ARD, IAD, RSLAD, and AKD · CIFAR-10 | Clean accuracy (%) | Hurt |
| Increase the inner-optimization perturbation limit (ablation) | PGD-AT, TRADES, ARD, IAD, RSLAD, and AKD · CIFAR-10 | AA accuracy (%) | Hurt |
| Use AdaIAD for self adversarial distillation instead of PGD-AT | ResNet-18 · CIFAR-10 self-distillation | AA accuracy (%): 47.69% to 50.19% | Helped |
Closest papers by meaning
- Out of Thin Air: Exploring Data-Free Adversarial Robustness Distillation (AAAI 2024 · similarity 0.90)
- Robust Distillation via Untargeted and Targeted Intermediate Adversarial Samples (CVPR 2024 · similarity 0.90)
- Indirect Gradient Matching for Adversarial Robust Distillation (ICLR 2025 · similarity 0.89)
- Learning Better Certified Models from Empirically-Robust Teachers (NeurIPS 2026 · similarity 0.89)
- Impartial Adversarial Distillation: Addressing Biased Data-Free Knowledge Distillation via Adaptive Constrained Optimization (AAAI 2024 · similarity 0.89)
- PeerAiD: Improving Adversarial Distillation from a Specialized Peer Tutor (CVPR 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.