Research map / Architectures and efficiency / Neural architecture search
Fitting the Search Space of Weight-sharing NAS with Graph Convolutional Networks
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 |
|---|---|---|---|
| fit sampled sub-network evaluations with a GCN before ranking candidates | weight-sharing NAS on ImageNet · ImageNet (ILSVRC2012), 8 architectures evaluated from scratch | Kendall-τ: 0.2143 -> 0.6429 | Helped |
| use GCN prediction to guide candidate selection | weight-sharing NAS · ImageNet (ILSVRC2012), 1,000 classes | top-1 accuracy (%): 75.2 -> 75.5 | Helped |
| add the SE module | One-Shot-NAS-GCN · ImageNet (ILSVRC2012) | top-1 accuracy (%): 75.5 -> 76.6 | Helped |
| preserve 6 candidates instead of 1 between search rounds (ablation) | One-Shot-NAS-GCN · ImageNet (ILSVRC2012) | top-1 accuracy (%): 75.34%±0.08% -> 75.47%±0.01% | Helped |
| increase sampled architectures per round from 2,000 to 5,000 (ablation) | One-Shot-NAS-GCN · ImageNet (ILSVRC2012), assigned similarity | test accuracy (%): 75.48 -> 75.47 | No clear effect |
| increase sampled architectures per round from 2,000 to 5,000 (ablation) | One-Shot-NAS-GCN · ImageNet (ILSVRC2012), measured similarity | test accuracy (%): 75.47 -> 75.45 | No clear effect |
| use measured similarity instead of assigned similarity (ablation) | One-Shot-NAS-GCN with 2,000 sampled architectures per round · ImageNet (ILSVRC2012) | test accuracy (%): 75.48 -> 75.47 | No clear effect |
| use measured similarity instead of assigned similarity (ablation) | One-Shot-NAS-GCN with 5,000 sampled architectures per round · ImageNet (ILSVRC2012) | test accuracy (%): 75.47 -> 75.45 | No clear effect |
| increase the graph edge Hamming-distance threshold to 2 (ablation) | GCN predictor · search-space performance prediction | GCN prediction accuracy | No clear effect |
| re-evaluate top-ranked GCN candidates with sub-network sampling (ablation) | One-Shot-NAS-GCN · ImageNet (ILSVRC2012) | top-1 accuracy (%): +0.1%--0.2% | Helped |
| increase super-network training from 150 to 450 epochs (ablation) | weight-sharing NAS · ImageNet (ILSVRC2012) | training accuracy (%): 86.17% -> 94.67% | Helped |
| increase super-network training from 150 to 450 epochs (ablation) | One-Shot-NAS-GCN search · ImageNet (ILSVRC2012) | final architecture quality | No clear effect |
Closest papers by meaning
- Distribution Consistent Neural Architecture Search (CVPR 2022 · similarity 0.92)
- Bridging the Gap between Sample-based and One-shot Neural Architecture Search with BONAS (NeurIPS 2020 · similarity 0.91)
- Landmark Regularization: Ranking Guided Super-Net Training in Neural Architecture Search (CVPR 2021 · similarity 0.90)
- Efficient Few-Shot Neural Architecture Search by Counting the Number of Nonlinear Functions (AAAI 2025 · similarity 0.90)
- Neural Architecture Search on ImageNet in Four GPU Hours: A Theoretically Inspired Perspective (ICLR 2021 · similarity 0.90)
- Can Weight Sharing Outperform Random Architecture Search? An Investigation With TuNAS (CVPR 2020 · 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.