Research map / Architectures and efficiency / Neural architecture search
Neural Architecture Search without Training
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 |
|---|---|---|---|
| Rank randomly sampled architectures by the untrained activation-overlap score and choose the highest-scoring candidate | NASWOT · NAS-Bench-201 and NATS-Bench SSS | accuracy | Helped |
| Increase the number of sampled architectures from N=10 to N=1000 | NASWOT · NAS-Bench-201 | CIFAR-10 test accuracy: 92.44 $\pm$ 1.13 -> 92.96 $\pm$ 0.81 | Helped |
| Increase the number of sampled architectures from N=10 to N=1000 | NASWOT · NATS-Bench SSS | CIFAR-10 test accuracy: 88.66 $\pm$ 0.90 -> 93.10 $\pm$ 0.31 | Helped |
| Use NASWOT scores to choose the initial population for REA | REA · NAS-Bench-201 / CIFAR-100 validation | accuracy: 71.81$\pm$1.12 -> 71.95 $\pm$ 0.99 | Helped |
| Use NASWOT scores to choose the initial population for REA | REA · NAS-Bench-201 / ImageNet-16-120 validation | accuracy: 45.15$\pm$0.89 -> 45.70 $\pm$ 1.05 | Helped |
| Use NASWOT scores to choose the initial population for REA | REA · NAS-Bench-101 / CIFAR-10 | accuracy: 93.87$\pm$0.22 -> 93.91$\pm$0.29 | No clear effect |
| Compute the score on different random CIFAR-100 mini-batches instead of a single fixed mini-batch (ablation) | NASWOT score · NAS-Bench-201 | architecture ranking robustness | No clear effect |
| Use normally distributed random inputs instead of image inputs (ablation) | NASWOT score · NAS-Bench-201 | score trend across architectures | No clear effect |
| Vary the network weight initialisation (ablation) | NASWOT score · NAS-Bench-201 | architecture ranking | No clear effect |
| Vary mini-batch size among 32, 64, 128, and 256 (ablation) | NASWOT score · NAS-Bench-201 | architecture ranking | No clear effect |
| Evaluate the score after training rather than only at initialisation (ablation) | NASWOT score · NAS-Bench-201 / CIFAR-10 | score ranking across networks | No clear effect |
Closest papers by meaning
- Neural Architecture Search on ImageNet in Four GPU Hours: A Theoretically Inspired Perspective (ICLR 2021 · similarity 0.92)
- Evaluating The Search Phase of Neural Architecture Search (ICLR 2020 · similarity 0.92)
- Unifying and Boosting Gradient-Based Training-Free Neural Architecture Search (NeurIPS 2022 · similarity 0.91)
- Per-Architecture Training-Free Metric Optimization for Neural Architecture Search (NeurIPS 2025 · similarity 0.90)
- Insights from the Use of Previously Unseen Neural Architecture Search Datasets (CVPR 2024 · similarity 0.90)
- Encodings for Prediction-based Neural Architecture Search (ICML 2024 · similarity 0.89)
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.