Research map / Label-efficient and robust learning / Knowledge distillation
RNAS-CL: Robust Neural Architecture Search by Cross-Layer Knowledge Distillation
Utkarsh Nath, Yan-Cheng Wang, Pavan K. Turaga, Yingzhen Yang
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 robust teacher with cross-layer distillation (RNAS-CL) vs standard training (ablation) | S3 search space · CIFAR-10 (no adversarial training) | PGD-20 accuracy: 17.69 -> 34.3 | Improved |
| add robust teacher with cross-layer distillation (RNAS-CL) vs standard training (ablation) | S5 search space · CIFAR-10 (no adversarial training) | PGD-20 accuracy: 18.44 -> 35.59 | Improved |
| add robust teacher with cross-layer distillation (RNAS-CL) vs standard training (ablation) | S7 search space · CIFAR-10 (no adversarial training) | PGD-20 accuracy: 19.3 -> 37.24 | Improved |
| add robust teacher with cross-layer distillation (RNAS-CL) vs standard training (ablation) | S3 search space · CIFAR-10 (no adversarial training) | clean accuracy: 89.92 -> 89.4 | No clear effect |
| use WRT-34 teacher instead of ResNet-18 teacher (ablation) | RNAS-CL-S3 · CIFAR-10 (no adversarial training) | PGD-20 accuracy: 26.35 -> 34.3 | Improved |
| use ResNet-50 teacher instead of ResNet-18 teacher (ablation) | RNAS-CL-S3 · CIFAR-10 (no adversarial training) | PGD-20 accuracy: 26.35 -> 35.76 | Improved |
| add TRADES adversarial retraining | RNAS-CL-M-WRT-34 · CIFAR-10 (no adversarial training) | PGD-20 accuracy: 39.84 -> 51.76 | Improved |
| add TRADES adversarial retraining | RNAS-CL-M-WRT-34 · CIFAR-10 (no adversarial training) | clean accuracy: 92.46 -> 87.29 | Hurt |
| add TRADES adversarial retraining | RNAS-CL-I-R-50 · ImageNet-100 | PGD-20 accuracy: 8.3 -> 32.44 | Improved |
| add TRADES adversarial retraining | RNAS-CL-I-R-50 · ImageNet-100 | clean accuracy: 85.98 -> 79.95 | Hurt |
| use RNAS-CL-WRT-50 instead of ResNet-50-R | adversarially trained ImageNet model · ImageNet (FastAT) | PGD-10 accuracy: 31.14 -> 33.30 | Improved |
| use RNAS-CL-S3-WRT-34 instead of MVV2-ARD | non-adversarially trained CIFAR-10 · CIFAR-10 | clean accuracy: 76.13 -> 89.4 | Improved |
Closest papers by meaning
- TRNAS: A Training-Free Robust Neural Architecture Search (ICCV 2025 · similarity 0.91)
- Robust NAS under adversarial training: benchmark, theory, and beyond (ICLR 2024 · similarity 0.91)
- Meta-prediction Model for Distillation-Aware NAS on Unseen Datasets (ICLR 2023 · similarity 0.90)
- Towards Accurate and Robust Architectures via Neural Architecture Search (CVPR 2024 · similarity 0.89)
- Robustifying and Boosting Training-Free Neural Architecture Search (ICLR 2024 · similarity 0.89)
- Insights from the Use of Previously Unseen Neural Architecture Search Datasets (CVPR 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.
