Research map / Architectures and efficiency / Neural architecture search
DDPNAS: Efficient Neural Architecture Search via Dynamic Distribution Pruning
Xiawu Zheng, Chenyi Yang, Shaokun Zhang, Yan Wang, Baochang Zhang, Yongjian Wu, Yunsheng Wu, Ling Shao, Rongrong Ji
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 dynamic operator pruning during search (ablation) | DDPNAS · CIFAR-10 | test error (%): 2.75 -> 2.59 | Improved |
| add dynamic operator pruning during search (ablation) | DDPNAS · CIFAR-10 | search cost (GPU days): 0.25 -> 0.075 | Improved |
| add dynamic pruning (ablation) | DDPNAS · GLUE dev | average score: 79.2 -> 80.4 | Improved |
| pruning step between 2 and 4 (ablation) | DDPNAS · CIFAR-10 | test error (%) | Improved |
| pruning step larger than 4 (ablation) | DDPNAS · CIFAR-10 | test error (%) | Hurt |
| change gamma within [0.7, 0.9] (ablation) | DDPNAS · CIFAR-10 | test error (%) | No clear effect |
| constrain network to N = 4 non-parameter ops instead of N = 2 (ablation) | DDPNAS · CIFAR-10 | test error (%): 2.72 -> 2.59 | Mixed |
| constrain height H = 1 with N = 4 (ablation) | DDPNAS · CIFAR-10 | test error (%): 2.59 -> 2.93 | Hurt |
| use DDPNAS over LayerDrop at 67M | BERT compression · GLUE dev (RTE) | accuracy: 65.2 -> 69.3 | Improved |
| use DDPNAS over LayerDrop at 67M | BERT compression · GLUE dev (CoLA) | Matthews corr.: 45.4 -> 54.1 | Improved |
| transfer DDPNAS backbone vs ESPNetV2 | semantic segmentation · Cityscapes | mIoU (%): 62.1 -> 71.6 | 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.
- DARTS — DARTS-: Robustly Stepping out of Performance Collapse Without Indicators (ICLR 2021) · better on 2 of 2 shared results
- XNAS — XNAS: Neural Architecture Search with Expert Advice (NeurIPS 2019) · 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
- DA-NAS: Data Adapted Pruning for Efficient Neural Architecture Search (ECCV 2020 · similarity 0.93)
- Neural Architecture Search on ImageNet in Four GPU Hours: A Theoretically Inspired Perspective (ICLR 2021 · similarity 0.92)
- MIGO-NAS: Towards Fast and Generalizable Neural Architecture Search (TPAMI 2021 · similarity 0.91)
- You Only Search Once: Single Shot Neural Architecture Search via Direct Sparse Optimization (TPAMI 2020 · similarity 0.91)
- DSNAS: Direct Neural Architecture Search Without Parameter Retraining (CVPR 2020 · similarity 0.91)
- ISTA-NAS: Efficient and Consistent Neural Architecture Search by Sparse Coding (NeurIPS 2020 · similarity 0.91)
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.
