Research map / Architectures and efficiency / Neural architecture search
Evaluating The Search Phase of Neural Architecture Search
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 |
|---|---|---|---|
| use NAS search rather than uniform random sampling | state-of-the-art NAS policies · PTB | validation perplexity: 59.88 $\pm$ 1.92; 60.61 $\pm$ 2.54; 61.99 $\pm$ 1.95; Random 60.13 $\pm$ 0.65 | No clear effect |
| use NAS search rather than uniform random sampling | state-of-the-art NAS policies · CIFAR-10 | top-1 accuracy: 96.79 $\pm$ 0.11; 96.62 $\pm$ 0.23; 96.86 $\pm$ 0.17; Random 96.44 $\pm$ 0.19 | No clear effect |
| use NAO rather than random sampling | NAO · RNN reduced space, 2 nodes | mean validation perplexity: 69.69 $\pm$ 2.44 -> 68.66 $\pm$ 2.50 | Helped |
| use NAO rather than random sampling | NAO · NASBench-101, 7 nodes | probability to surpass random: 0.62 | Helped |
| remove weight sharing (ablation) | NAO · NASBench-101, 7 nodes | best test accuracy: 93.33 -> 94.11 | Helped |
| remove weight sharing (ablation) | ENAS · NASBench-101, 7 nodes | best test accuracy: 92.54 -> 94.04 | Helped |
| remove weight sharing (ablation) | NAO · NASBench-101, 7 nodes | probability to surpass random: 0.62 to 0.92 | Helped |
| remove weight sharing (ablation) | ENAS · NASBench-101, 7 nodes | probability to surpass random: 0.07 to 0.90 | Helped |
| train architectures with increasing amounts of shared weights (ablation) | weight-sharing ranking · NASBench-101, 6 nodes | Kendall Tau: 0.67; 0.33; -0.33; 0.0 | Mixed |
| increase search-space size (ablation) | weight-sharing methods · NASBench-101, 3-to-7-node spaces | Kendall Tau: 0.441, 0.314, 0.214, 0.195 | Hurt |
| use NAO instead of random search | NAO · CIFAR-10, 7-node DARTS search space | best test top-1 accuracy: 96.74 -> 97.10 | Helped |
Closest papers by meaning
- NAS evaluation is frustratingly hard (ICLR 2020 · similarity 0.93)
- Neural Architecture Search on ImageNet in Four GPU Hours: A Theoretically Inspired Perspective (ICLR 2021 · similarity 0.93)
- Neural Architecture Generator Optimization (NeurIPS 2020 · similarity 0.92)
- Neural Architecture Search With Representation Mutual Information (CVPR 2022 · similarity 0.92)
- Insights from the Use of Previously Unseen Neural Architecture Search Datasets (CVPR 2024 · similarity 0.92)
- einspace: Searching for Neural Architectures from Fundamental Operations (NeurIPS 2024 · similarity 0.90)
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.