Research map / Architectures and efficiency / Neural architecture search
Can Weight Sharing Outperform Random Architecture Search? An Investigation With TuNAS
Check 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 TuNAS weight-sharing search instead of random search (N=50) | ProxylessNAS search space · ImageNet | test accuracy (%): 73.2 \pm 0.3 -> 75.0 \pm 0.1 | Helped |
| use TuNAS weight-sharing search instead of random search (N=50) | ProxylessNAS-Enlarged search space · ImageNet | test accuracy (%): 73.2 \pm 0.3 -> 75.3 \pm 0.2 | Helped |
| use TuNAS weight-sharing search instead of random search (N=50) | MobileNetV3-Like search space · ImageNet | test accuracy (%): 73.5 \pm 0.2 -> 75.4 \pm 0.1 | Helped |
| search the detection backbone with TuNAS | MobileNetV3-Like space with SSDLite · COCO test-dev | mAP: 22.0 -> 22.5 | Helped |
| search output filter sizes rather than use the every-stride-2 doubling heuristic (ablation) | ProxylessNAS search space · ImageNet | test accuracy (%): 73.5 \pm 0.2 -> 75.0 \pm 0.1 | Helped |
| search output filter sizes rather than double filters at every block (ablation) | ProxylessNAS search space · ImageNet | test accuracy (%): 74.0 \pm 0.2 -> 75.0 \pm 0.1 | Helped |
| use the absolute value reward instead of the MnasNet soft reward (ablation) | TuNAS on ProxylessNAS search space · ImageNet | test accuracy (%): 74.8 \pm 0.3 -> 75.0 \pm 0.1 | No clear effect |
| add op and filter warmup (ablation) | TuNAS on ProxylessNAS search space · ImageNet | test accuracy (%): 74.7 \pm 0.1 -> 75.0 \pm 0.1 | Helped |
| add op and filter warmup (ablation) | TuNAS on ProxylessNAS-Enlarged search space · ImageNet | test accuracy (%): 74.6 \pm 0.2 -> 75.3 \pm 0.2 | Helped |
| add op and filter warmup (ablation) | TuNAS on MobileNetV3-Like search space · ImageNet | test accuracy (%): 75.0 \pm 0.1 -> 75.2 \pm 0.2 | Helped |
| use aggressive weight sharing (ablation) | TuNAS on ProxylessNAS search space · ImageNet | test accuracy (%): 75.0 \pm 0.1 -> 75.0 \pm 0.1 | No clear effect |
| increase search training from 90 to 360 epochs (ablation) | TuNAS on MobileNetV3-Like search space · ImageNet | test accuracy (%): 75.2 \pm 0.2 -> 75.4 \pm 0.1 | Helped |
Closest papers by meaning
- Evaluating The Search Phase of Neural Architecture Search (ICLR 2020 · similarity 0.90)
- Vision-Oriented Lightweight Neural Architecture Search with Budget-Adaptive Evaluation (CVPR 2026 · similarity 0.90)
- Fitting the Search Space of Weight-sharing NAS with Graph Convolutional Networks (AAAI 2021 · similarity 0.90)
- Cream of the Crop: Distilling Prioritized Paths For One-Shot Neural Architecture Search (NeurIPS 2020 · similarity 0.90)
- Bridging the Gap between Sample-based and One-shot Neural Architecture Search with BONAS (NeurIPS 2020 · similarity 0.89)
- Hierarchical Representations for Efficient Architecture Search (ICLR 2018 · 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.