Research map / Adversarial robustness and security / Certified and empirical robustness
GCMRD: Global Consistency Multi-teacher Robustness Distillation
Yuhang Zhou, Zhongyun Hua, Rushi Lan, Qing Liao, Wei Jiang
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 |
|---|---|---|---|
| add MCIM (dual-teacher inner maximization) (ablation) | CIARD · CIFAR-10 ResNet-18 | clean accuracy: 88.87% -> 86.42% | Hurt |
| add MCIM (dual-teacher inner maximization) (ablation) | CIARD · CIFAR-10 ResNet-18 PGD-TRADES | robust accuracy: 54.46% -> 54.84% | Improved |
| add MCIM (dual-teacher inner maximization) (ablation) | CIARD · CIFAR-10 ResNet-18 CW | robust accuracy: 50.61% -> 51.30% | Improved |
| add MCIM (dual-teacher inner maximization) (ablation) | CIARD · CIFAR-100 ResNet-18 | clean accuracy: 65.73% -> 63.27% | Hurt |
| add MCIM (dual-teacher inner maximization) (ablation) | CIARD · CIFAR-100 ResNet-18 PGD-SAT | robust accuracy: 28.05% -> 28.74% | Improved |
| add FRR (full-repulsion regularization) (ablation) | CIARD · CIFAR-10 ResNet-18 | clean accuracy: 88.87% -> 87.06% | Hurt |
| add FRR (full-repulsion regularization) (ablation) | CIARD · CIFAR-10 ResNet-18 PGD-SAT | robust accuracy: 51.70% -> 52.75% | Improved |
| add FRR (full-repulsion regularization) (ablation) | CIARD · CIFAR-100 ResNet-18 CW | robust accuracy: 24.43% -> 25.72% | Improved |
| add CFT (natural-teacher guided robust teacher fine-tuning) (ablation) | CIARD · CIFAR-10 ResNet-18 | clean accuracy: 88.87% -> 89.37% | Improved |
| add CFT (natural-teacher guided robust teacher fine-tuning) (ablation) | CIARD · CIFAR-10 ResNet-18 PGD-SAT | robust accuracy: 51.70% -> 51.90% | No clear effect |
| add all three modules (full GCMRD) (ablation) | CIARD · CIFAR-10 ResNet-18 | clean accuracy: 88.87% -> 89.26% | Improved |
| add all three modules (full GCMRD) (ablation) | CIARD · CIFAR-100 ResNet-18 CW | robust accuracy: 24.43% -> 25.89% | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- CIARD — CIARD: Cyclic Iterative Adversarial Robustness Distillation (ICCV 2025) · better on 11 of 11 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- CIARD: Cyclic Iterative Adversarial Robustness Distillation (ICCV 2025 · similarity 0.92)
- Out of Thin Air: Exploring Data-Free Adversarial Robustness Distillation (AAAI 2024 · similarity 0.89)
- 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 0.88)
- Indirect Gradient Matching for Adversarial Robust Distillation (ICLR 2025 · similarity 0.87)
- Improving Adversarial Robust Fairness via Anti-Bias Soft Label Distillation (NeurIPS 2024 · 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.
