Research map / Architectures and efficiency / Neural architecture search
An Evaluation of Zero-Cost Proxies - from Neural Architecture Performance Prediction to Model Robustness
Jovita Lukasik, Michael Moeller, Margret Keuper
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 all zero-cost proxies instead of only the most important one (ablation) | random forest predictor · CIFAR-10 PGD robust accuracy | test R2: -0.31 -> 0.65 | Improved |
| use all zero-cost proxies instead of only the most important one (ablation) | random forest predictor · CIFAR-10 clean accuracy | test R2: 0.84 -> 0.97 | Improved |
| use all zero-cost proxies instead of only the most important one (ablation) | random forest predictor · ImageNet16-120 FGSM accuracy | test R2: -0.003 -> 0.65 | Improved |
| use all zero-cost proxies instead of only the most important one (ablation) | random forest predictor · CIFAR-100 clean accuracy | test R2: 0.74 -> 0.96 | Improved |
| add the most important proxy to the feature set (ablation) | random forest multi-objective · CIFAR-10 Clean-FGSM | test R2: 0.92 -> 0.92 | No clear effect |
| add the most important proxy to the feature set (ablation) | random forest multi-objective · CIFAR-10 Clean-PGD | test R2: 0.79 -> 0.81 | Improved |
| add the most important proxy to the feature set (ablation) | random forest multi-objective · ImageNet16-120 Clean-FGSM | test R2: 0.72 -> 0.78 | Improved |
| add the most important proxy to the feature set (ablation) | random forest multi-objective · CIFAR-100 Clean-PGD | test R2: 0.83 -> 0.83 | No clear effect |
| use stronger attacks (PGD/APGD instead of FGSM) | Kendall tau of ZCPs with robust accuracy · CIFAR-10 | rank correlation | Hurt |
Closest papers by meaning
- Zero-Cost Proxies for Lightweight NAS (ICLR 2021 · similarity 0.93)
- NAS-Bench-Suite-Zero: Accelerating Research on Zero Cost Proxies (NeurIPS 2022 · similarity 0.93)
- ParZC: Parametric Zero-Cost Proxies for Efficient NAS (AAAI 2025 · similarity 0.91)
- Surprisingly Strong Performance Prediction with Neural Graph Features (ICML 2024 · similarity 0.90)
- AZ-NAS: Assembling Zero-Cost Proxies for Network Architecture Search (CVPR 2024 · similarity 0.89)
- NEAR: A Training-Free Pre-Estimator of Machine Learning Model Performance (ICLR 2025 · 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.
