Research map / Architectures and efficiency / Neural architecture search
Pi-NAS: Improving Neural Architecture Search by Reducing Supernet Training Consistency Shift
Jiefeng Peng, Jiqi Zhang, Changlin Li, Guangrun Wang, Xiaodan Liang, Liang Lin
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 |
|---|---|---|---|
| add cross-path learning and nontrivial mean teacher to SPOS (full Pi-NAS) (ablation) | SPOS supernet · Ranking correlation (classification) | Kendall's Tau: 0.19 -> 0.79 | Improved |
| add mean teacher to the supernet-Pi model (ablation) | Pi-NAS · Ranking correlation (classification) | Kendall's Tau: 0.48 -> 0.79 | Improved |
| add cross-path learning (ablation) | Pi-NAS · Ranking correlation (classification) | Kendall's Tau: 0.14 -> 0.79 | Improved |
| add downsample-sharing (ablation) | Pi-NAS · Ranking correlation (classification) | Kendall's Tau: 0.40 -> 0.79 | Improved |
| use nontrivial mean teacher with negative targets instead of trivial mean teacher (ablation) | Pi-NAS · Supernet training | Kendall's Tau: collision -> 0.79 | Improved |
| replace SPOS/ResNeSt-50 with Pi-NAS-cls found architecture | ResNeSt-50 search space · ImageNet val | top-1 accuracy: 81.1 -> 81.6 | Improved |
| Pi-NAS vs SPOS supernet for architecture search | searched models · ImageNet val | top-1 accuracy: 81.04 -> 81.6 | Improved |
| use Pi-NAS-trans backbone instead of ResNeSt-50 | Mask R-CNN · COCO 2017 val | AP box: 42.81 -> 44.11 | Improved |
| use Pi-NAS-trans backbone instead of ResNeSt-50 | DeeplabV3 · ADE20K val | mIoU: 45.18 -> 45.49 | Improved |
| use Pi-NAS-trans backbone instead of ResNeSt-50 | DeeplabV3 · Cityscapes val | mIoU: 80.08 -> 80.40 | Improved |
| use Pi-NAS ranking for instance segmentation transfer vs SPOS | supernet · Ranking correlation (instance segmentation) | Kendall's Tau: 0.18 -> 0.51 | Improved |
| apply Pi-NAS on NAS-Bench-201 instead of SPOS | supernet · NAS-Bench-201 CIFAR-10 | test accuracy: 93.57 -> 93.83 | 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.
- FairNAS — FairNAS: Rethinking Evaluation Fairness of Weight Sharing Neural Architecture Search (ICCV 2021) · better on 2 of 2 shared results
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
- Boosting Order-Preserving and Transferability for Neural Architecture Search: a Joint Architecture Refined Search and Fine-tuning Approach (CVPR 2024 · similarity 0.91)
- Neural Architecture Search on ImageNet in Four GPU Hours: A Theoretically Inspired Perspective (ICLR 2021 · similarity 0.91)
- Subnet-Aware Dynamic Supernet Training for Neural Architecture Search (CVPR 2025 · similarity 0.90)
- NetAdaptV2: Efficient Neural Architecture Search With Fast Super-Network Training and Architecture Optimization (CVPR 2021 · similarity 0.90)
- Distribution Consistent Neural Architecture Search (CVPR 2022 · similarity 0.90)
- Efficient Few-Shot Neural Architecture Search by Counting the Number of Nonlinear Functions (AAAI 2025 · 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.
