Research map / Architectures and efficiency / Neural architecture search
Are Labels Necessary for Neural Architecture Search?
Chenxi Liu, Piotr Dollár, Kaiming He, Ross Girshick, Alan Yuille, Saining Xie
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 with rotation prediction instead of labels | DARTS · ImageNet-1K val (search IN1K) | top-1 accuracy (%): 76.3 -> 75.8 | No clear effect |
| search with labels replaced by jigsaw (UnNAS) | DARTS · ImageNet-1K val (search IN22K) | top-1 accuracy (%): 75.9 -> 75.9 | No clear effect |
| search with colorization on Cityscapes instead of segmentation labels | DARTS · ImageNet-1K val (search Cityscapes) | top-1 accuracy (%): 75.8 -> 75.2 | Hurt |
| search with jigsaw instead of segmentation labels on Cityscapes | DARTS · Cityscapes val mIoU | mIoU (%): 72.4 -> 74.1 | Improved |
| search with unsupervised objectives instead of classification labels on IN22K | DARTS · Cityscapes val mIoU | mIoU (%): 72.4 -> 73.6 | Improved |
| search with colorization instead of classification labels on IN1K | DARTS · Cityscapes val mIoU | mIoU (%): 73.6 -> 72.2 | Hurt |
| search on IN22K instead of IN1K (10x more images) | NAS/UnNAS DARTS · ImageNet-1K val | top-1 accuracy (%): 76.3 -> 75.9 | No clear effect |
| search with UnNAS on Cityscapes jigsaw instead of CIFAR-10 supervised proxy architecture | DARTS · Cityscapes val mIoU | mIoU (%): 72.6 -> 74.1 | Improved |
| use unsupervised pretext accuracy as proxy instead of CIFAR-10 supervised accuracy for ranking | DARTS search space · rank correlation with ImageNet accuracy | Spearman rho: 0.90 -> ~0.8 | No clear effect |
Closest papers by meaning
- Does Unsupervised Architecture Representation Learning Help Neural Architecture Search? (NeurIPS 2020 · similarity 0.90)
- Data-Free Neural Architecture Search via Recursive Label Calibration (ECCV 2022 · similarity 0.89)
- Neural Architecture Search With Random Labels (CVPR 2021 · similarity 0.88)
- Neural Architecture Search: A Survey (JMLR 2019 · similarity 0.87)
- Insights from the Use of Previously Unseen Neural Architecture Search Datasets (CVPR 2024 · similarity 0.86)
- Building Optimal Neural Architectures using Interpretable Knowledge (CVPR 2024 · similarity 0.85)
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.
