Research map / Architectures and efficiency / Neural architecture search
Zero-Cost Proxies for Lightweight NAS
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 |
|---|---|---|---|
| combine synflow, jacob_cov, and snip rankings with majority voting | zero-cost NAS proxy scoring · NAS-Bench-201 CIFAR-10 | Spearman ρ: 0.74 -> 0.82 | Helped |
| use synflow instead of EcoNAS r8c4e20 | NAS-Bench-201 model ranking · NAS-Bench-201 CIFAR-10 | Spearman ρ: 0.61 -> 0.74 | Helped |
| use synflow warmup for random search | RAND · NAS-Bench-201 CIFAR-100 | models to reach 73.5% accuracy: 1000+ -> 8 | Helped |
| use synflow warmup for binary predictor search | binary predictor · NAS-Bench-101 CIFAR-10 | sample efficiency | No clear effect |
| increase RL synflow warmup to 15k points | reinforcement learning search · NAS-Bench-201 | search performance | Hurt |
| use synflow warmup with random search | RAND · NAS-Bench-ASR TIMIT | models to reach 21.3% PER: 1000+ -> 68 | Helped |
| use synflow warmup with aging evolution | AE · NAS-Bench-ASR TIMIT | models to reach 21.3% PER: 138 -> 87 | Helped |
| replace synflow with jacob_cov for ranking unrelated architectures | zero-cost scoring · PyTorchCV models | Spearman ρ | Hurt |
| initialize biases to zero (ablation) | fisher proxy · NAS-Bench-201 CIFAR-10 | Spearman ρ: 0.357 -> 0.509 | Helped |
| increase grasp minibatch from 32 to 512 samples (ablation) | grasp proxy · NAS-Bench-201 CIFAR-10 | Spearman ρ: 0.511 -> 0.381 | Hurt |
| increase zero-cost warmup to 15k points for RL (ablation) | reinforcement learning search · NAS-Bench-201 | search performance | Hurt |
| add synflow warmup to binary predictor search (ablation) | binary predictor · NAS-Bench-101 CIFAR-10 | sample efficiency | No clear effect |
Closest papers by meaning
- AZ-NAS: Assembling Zero-Cost Proxies for Network Architecture Search (CVPR 2024 · similarity 0.91)
- ParZC: Parametric Zero-Cost Proxies for Efficient NAS (AAAI 2025 · similarity 0.91)
- NEAR: A Training-Free Pre-Estimator of Machine Learning Model Performance (ICLR 2025 · similarity 0.91)
- Surprisingly Strong Performance Prediction with Neural Graph Features (ICML 2024 · similarity 0.87)
- Encodings for Prediction-based Neural Architecture Search (ICML 2024 · similarity 0.87)
- SWAP-NAS: Sample-Wise Activation Patterns for Ultra-fast NAS (ICLR 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.