Research map / Architectures and efficiency / Neural architecture search
Evolving Search Space for Neural Architecture Search
Yuanzheng Ci, Chen Lin, Ming Sun, Boyu Chen, Hongwen Zhang, Wanli Ouyang
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (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 the 27 OPs space with NSE instead of ProxylessNAS | ProxylessNAS · ImageNet (27 OPs space) | top-1 accuracy (%): 74.5 -> 75.0 | Improved |
| search the 27 OPs space with NSE instead of One-Shot | One-Shot · ImageNet (27 OPs space) | top-1 accuracy (%): 73.5 -> 75.0 | Improved |
| enlarge ProxylessNAS search space from manual 6 OPs to 27 OPs | ProxylessNAS · ImageNet | top-1 accuracy (%): 74.6 -> 74.5 | No clear effect |
| use random 6 OPs subsets instead of manual 6 OPs space | ProxylessNAS · ImageNet | top-1 accuracy (%): 74.6 -> 74.2 | Hurt |
| add a second search space after the 27 OPs space (ablation) | NSE · ImageNet (FLOPs-constrained) | top-1 accuracy (%): 75.3 -> 75.5 | Improved |
| add SE and swish modules to the searched network | NSENet · ImageNet (FLOPs-constrained) | top-1 accuracy (%): 75.5 -> 77.3 | Improved |
| use NSE with latency constraint instead of PC-NAS-L in the same space | PC-NAS-L · ImageNet (latency-constrained) | top-1 accuracy (%): 77.5 -> 77.9, 10.3 ms -> 8.9 ms | Improved |
| swap EfficientNet-B0 backbone for NSENet | EfficientDet-D0 · COCO | mAP: 33.8 -> 34.5 | Improved |
| swap MobileNetV2 backbone for NSENet | EfficientDet-D0 · COCO | mAP: 32.7 -> 34.5 | Improved |
| enlarge initial layer-wise search space from K=5 to K=9 without evolution (ablation) | NSE with simplification · ImageNet (retrain accuracy) | top-1 accuracy (%) | Hurt |
| add Lock and Rehearse (ablation) | NSE (K=9) · ImageNet (retrain accuracy) | top-1 accuracy (%) | Improved |
| add search space simplification and Pareto aggregation (ablation) | NSE round 1 (K=5) · ImageNet (50-epoch retrain accuracy) | top-1 accuracy (%) | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- DARTS — DARTS-: Robustly Stepping out of Performance Collapse Without Indicators (ICLR 2021) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- einspace: Searching for Neural Architectures from Fundamental Operations (NeurIPS 2024 · similarity 0.92)
- Neural Architecture Generator Optimization (NeurIPS 2020 · similarity 0.92)
- Neural Architecture Search on ImageNet in Four GPU Hours: A Theoretically Inspired Perspective (ICLR 2021 · similarity 0.92)
- Understanding and Accelerating Neural Architecture Search With Training-Free and Theory-Grounded Metrics (TPAMI 2023 · similarity 0.91)
- Towards Neural Architecture Search through Hierarchical Generative Modeling (ICML 2024 · similarity 0.91)
- EG-NAS: Neural Architecture Search with Fast Evolutionary Exploration (AAAI 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.
