Research map / Adversarial robustness and security / Certified and empirical robustness
Accelerating Certified Robustness Training via Knowledge Transfer
Pratik Vaishnavi, Kevin Eykholt, Amir Rahmati
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 |
|---|---|---|---|
| train student with knowledge transfer instead of SmoothMix | RegNet · CIFAR-10, sigma=0.25 | ACR: 0.538 -> 0.564 | Improved |
| train student with knowledge transfer instead of SmoothMix | ResNeXt · CIFAR-10, sigma=0.25 | ACR: 0.516 -> 0.558 | Improved |
| train student with knowledge transfer instead of SmoothMix | DLA · CIFAR-10, sigma=0.25 | certified accuracy at radius 0.00: 77.72 -> 77.31 | No clear effect |
| use knowledge transfer instead of SmoothMix for training speed | ResNeXt/DLA/RegNet · CIFAR-10 | average per-epoch speedup: 10.75 × | Improved |
| use knowledge transfer instead of SmoothMix for total training time | ResNeXt/DLA/RegNet · CIFAR-10 | total training time: 8.06 × | Improved |
| use knowledge transfer instead of SmoothMix for ResNeXt | ResNeXt · CIFAR-10 | total time (h): 45.21 -> 4.80 | Improved |
| apply knowledge transfer recursively (chain length 1 to 3) (ablation) | KT framework · CIFAR-10, sigma=0.25 | ACR: 0.558 -> 0.559 | No clear effect |
| use a small ResNet-S teacher trained with SmoothMix to train ResNet-L (ablation) | KT framework · CIFAR-10 | ACR: 0.550 -> 0.540 | No clear effect |
| use a small ResNet-S teacher to train ResNet-L (ablation) | KT framework · CIFAR-10 | total time (h): 18.98 -> 10.07 | Improved |
| swap teacher training method from SmoothMix to MACER (ablation) | KT framework · CIFAR-10 student ACR | ACR: 0.540 -> 0.528 | Hurt |
| swap teacher training method from SmoothMix to SmoothAdv (ablation) | KT framework · CIFAR-10 student ACR | ACR: 0.540 -> 0.519 | Hurt |
| transfer robustness on ImageNet versus training student directly | KT framework · ImageNet, SmoothAdv | ACR: 0.820 -> 0.684 | Hurt |
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.
- SmoothMix — SmoothMix: Training Confidence-calibrated Smoothed Classifiers for Certified Robustness (NeurIPS 2021) · better on 15 of 15 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
- Learning Better Certified Models from Empirically-Robust Teachers (NeurIPS 2026 · similarity 0.90)
- On the Scalability of Certified Adversarial Robustness with Generated Data (NeurIPS 2024 · similarity 0.87)
- CTBench: A Library and Benchmark for Certified Training (ICML 2025 · similarity 0.87)
- Sparse Transfer Learning Accelerates and Enhances Certified Robustness: A Comprehensive Study (AAAI 2025 · similarity 0.86)
- Halt Fast! Early Stopping for Certified Robustness (NeurIPS 2026 · similarity 0.86)
- Boosting Adversarial Robustness with CLAT: Criticality Leveraged Adversarial Training (ICML 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.
