Research map / Architectures and efficiency / Neural architecture search
NASI: Label- and Data-agnostic Neural Architecture Search at Initialization
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 |
|---|---|---|---|
| search at initialization without model training | NASI on CIFAR-10 · CIFAR-10 | search cost (GPU hours): 1.2 -> 0.24 | Helped |
| select architectures using NASI-FIX rather than the strongest listed baseline | CIFAR-100 architecture search · CIFAR-100 test | test error (%): 17.51$\pm$0.25 -> 16.12$\pm$0.38 | Helped |
| select architectures using NASI rather than P-DARTS | CIFAR-10 architecture search · CIFAR-10 test | test error (%): 2.50 -> 2.79$\pm$0.07 | Hurt |
| use random labels instead of true labels (ablation) | NASI · NAS-Bench-1Shot1 S1, CIFAR-10 | test error (%): 7.3$\pm$1.1 -> 6.8$\pm$0.3 | No clear effect |
| use random data instead of true data (ablation) | NASI · NAS-Bench-1Shot1 S1, CIFAR-10 | test error (%): 7.3$\pm$1.1 -> 6.6$\pm$0.2 | No clear effect |
| use random labels instead of true labels (ablation) | NASI · NAS-Bench-1Shot1 S2, CIFAR-10 | test error (%): 7.2$\pm$0.4 -> 7.0$\pm$0.4 | No clear effect |
| use random data instead of true data (ablation) | NASI · NAS-Bench-1Shot1 S2, CIFAR-10 | test error (%): 7.2$\pm$0.4 -> 7.5$\pm$0.7 | No clear effect |
| use random labels instead of true labels (ablation) | NASI · NAS-Bench-1Shot1 S3, CIFAR-10 | test error (%): 7.2$\pm$0.6 -> 7.5$\pm$1.4 | No clear effect |
| use random data instead of true data (ablation) | NASI · NAS-Bench-1Shot1 S3, CIFAR-10 | test error (%): 7.2$\pm$0.6 -> 7.3$\pm$0.9 | No clear effect |
| use random labels instead of true labels (ablation) | NASI · DARTS search space, CIFAR-10 | test error (%): 2.95$\pm$0.13 -> 2.90$\pm$0.12 | No clear effect |
| use random data instead of true data (ablation) | NASI · DARTS search space, CIFAR-10 | test error (%): 2.95$\pm$0.13 -> 2.97$\pm$0.10 | No clear effect |
Closest papers by meaning
- Neural Architecture Search With Representation Mutual Information (CVPR 2022 · similarity 0.91)
- NAS-Bench-201: Extending the Scope of Reproducible Neural Architecture Search (ICLR 2020 · similarity 0.90)
- Neural Architecture Search on ImageNet in Four GPU Hours: A Theoretically Inspired Perspective (ICLR 2021 · similarity 0.90)
- Evaluating The Search Phase of Neural Architecture Search (ICLR 2020 · similarity 0.90)
- Insights from the Use of Previously Unseen Neural Architecture Search Datasets (CVPR 2024 · similarity 0.89)
- einspace: Searching for Neural Architectures from Fundamental Operations (NeurIPS 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.