Research map / Architectures and efficiency / Neural architecture search
DA-NAS: Data Adapted Pruning for Efficient Neural Architecture Search
Xiyang Dai, Dongdong Chen, Mengchen Liu, Yinpeng Chen, Lu Yuan
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 |
|---|---|---|---|
| start with easy classes then all classes instead of searching on all classes directly (ablation) | DA-NAS · ImageNet val (Shuffle space) | top-1 accuracy: 73.8 -> 74.4 | Improved |
| start with easy classes instead of hard classes (ablation) | DA-NAS data scheduling · ImageNet val (Shuffle space) | top-1 accuracy: 74.1 -> 74.4 | Improved |
| progress from a small subset to all classes instead of a small subset only (ablation) | DA-NAS data scheduling · ImageNet val (Shuffle space) | top-1 accuracy: 73.2 -> 74.4 | Improved |
| add easy-to-all data scheduling without pruning (ablation) | DA-NAS · ImageNet val (Shuffle space) | top-1 accuracy: 73.8 -> 74.2 | Improved |
| add search space pruning without data scheduling (ablation) | DA-NAS · ImageNet val (Shuffle space) | top-1 accuracy: 73.8 -> 74.2 | Improved |
| add pruning to data scheduling (ablation) | DA-NAS · ImageNet val (Shuffle space) | top-1 accuracy: 74.2 -> 74.4 | Improved |
| add data scheduling and pruning together (ablation) | DA-NAS · ImageNet val (Shuffle space) | search cost (GPU hours): 307 -> 87 | Improved |
| enlarge search space to Shuffle+ (ablation) | DA-NAS · ImageNet val (Shuffle space) | top-1 accuracy: 74.4 -> 74.3 | No clear effect |
| enlarge search space to Shuffle+Mobile (ablation) | DA-NAS · ImageNet val (Shuffle space) | top-1 accuracy: 74.4 -> 74.3 | No clear effect |
| use DA-NAS instead of SinglePath in the Shuffle space | SinglePath · ImageNet val (Shuffle space) | search cost (GPU hours): 142 -> 87 | Improved |
| use DA-NAS instead of Proxyless-G in the Mobile space | Proxyless-G · ImageNet val (Mobile space) | search cost (GPU hours): 399 -> 138 | Improved |
| use DA-NAS under half the search time in Shuffle+ | SinglePath · ImageNet val (Shuffle+ space, 0.5x search time) | top-1 accuracy: 72.4 -> 73.3 | Improved |
Closest papers by meaning
- Neural Architecture Search on ImageNet in Four GPU Hours: A Theoretically Inspired Perspective (ICLR 2021 · similarity 0.94)
- DDPNAS: Efficient Neural Architecture Search via Dynamic Distribution Pruning (IJCV 2023 · similarity 0.93)
- Rapid Neural Architecture Search by Learning to Generate Graphs from Datasets (ICLR 2021 · similarity 0.92)
- Block-Wisely Supervised Neural Architecture Search With Knowledge Distillation (CVPR 2020 · similarity 0.92)
- Understanding and Accelerating Neural Architecture Search With Training-Free and Theory-Grounded Metrics (TPAMI 2023 · similarity 0.92)
- ISTA-NAS: Efficient and Consistent Neural Architecture Search by Sparse Coding (NeurIPS 2020 · similarity 0.92)
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.
