Research map / Architectures and efficiency / Neural architecture search
BN-NAS: Neural Architecture Search With Batch Normalization
Boyu Chen, Peixia Li, Baopu Li, Chen Lin, Chuming Li, Ming Sun, Junjie Yan, 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 |
|---|---|---|---|
| use BN-indicator instead of validation accuracy to rank subnets and train only 10 epochs | SPOS · ImageNet val | top-1 accuracy (%): 75.73 -> 75.67 | No clear effect |
| use BN-indicator and 15 epochs of BN-only supernet training | FairNAS · ImageNet val | top-1 accuracy (%): 74.07 -> 74.12 | No clear effect |
| replace validation-accuracy subnet search by BN-indicator | SPOS · ImageNet search cost | subnet search cost: ~1 GPU day -> 0.14s on CPU | Improved |
| reduce supernet epochs and train only BN layers | SPOS · ImageNet supernet training | epochs: 100 -> 10 | Improved |
| train only BN parameters instead of all parameters (ablation) | SPOS supernet · ImageNet supernet training | training time: about 20% speedup | Improved |
| use BN-indicator instead of accuracy indicator with 100 epochs of all-parameter training (ablation) | SPOS · ImageNet | searched subnet accuracy: All/BN/100 vs All/Acc/100 similar | No clear effect |
| use BN-indicator instead of accuracy indicator with 30 epochs (ablation) | SPOS · ImageNet | searched subnet accuracy: All/BN/30 better than All/Acc/30 | Improved |
| use accuracy indicator with 30 epoch supernet versus random baseline (ablation) | SPOS · ImageNet | searched subnet accuracy: +0.1% over random baseline | No clear effect |
| BN-indicator with all-parameter training cut to 10 epochs (ablation) | SPOS · ImageNet | searched subnet accuracy: 10 epochs All/BN fails to keep performance | Hurt |
| train only BN parameters for 10 epochs (ablation) | BN-NAS · ImageNet | searched subnet accuracy: BN/BN/10 recovers performance | Improved |
| BN-indicator versus accuracy on rank correlation with retrain accuracy (ablation) | SPOS · CIFAR10 | Kendall Tau: similar Kendall Tau | No clear effect |
| add SE modules to BN-NAS SPOS model | SPOS(Ours) · ImageNet val | top-1 accuracy (%): 75.67 -> 76.78 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- FairNAS — FairNAS: Rethinking Evaluation Fairness of Weight Sharing Neural Architecture Search (ICCV 2021) · better on 3 of 3 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
- Understanding and Accelerating Neural Architecture Search With Training-Free and Theory-Grounded Metrics (TPAMI 2023 · similarity 0.91)
- Subnet-Aware Dynamic Supernet Training for Neural Architecture Search (CVPR 2025 · similarity 0.91)
- NetAdaptV2: Efficient Neural Architecture Search With Fast Super-Network Training and Architecture Optimization (CVPR 2021 · similarity 0.91)
- NAS-Bench-201: Extending the Scope of Reproducible Neural Architecture Search (ICLR 2020 · similarity 0.91)
- Boosting Order-Preserving and Transferability for Neural Architecture Search: a Joint Architecture Refined Search and Fine-tuning Approach (CVPR 2024 · similarity 0.90)
- Efficient Few-Shot Neural Architecture Search by Counting the Number of Nonlinear Functions (AAAI 2025 · 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.
