Research map / Architectures and efficiency / Neural architecture search
DCP–NAS: Discrepant Child–Parent Neural Architecture Search for 1-bit CNNs
Yanjing Li, Sheng Hua Xu, Xianbin Cao, Li’an Zhuo, Baochang Zhang, Tian Wang, Guodong Guo
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 tangent direction constraint (ablation) | DCP-NAS-L · ImageNet | top-1 accuracy: 66.7 -> 68.3 | Improved |
| add reconstruction error term (ablation) | DCP-NAS-L · ImageNet | top-1 accuracy: 66.7 -> 68.2 | Improved |
| add both tangent constraint and reconstruction error (ablation) | DCP-NAS-L · ImageNet | top-1 accuracy: 66.7 -> 72.4 | Improved |
| add GGN approximation of the Hessian (ablation) | DCP-NAS-L · ImageNet | search cost (GPU days): 27.9 -> 2.9 | Improved |
| add decoupled optimization (ablation) | DCP-NAS-L (tangent propagation + GGN) · ImageNet | top-1 accuracy: 71.2 -> 72.4 | Improved |
| add decoupled optimization (ablation) | DARTS · ImageNet | top-1 accuracy: 73.1 -> 73.8 | Improved |
| add decoupled optimization (ablation) | PC-DARTS · ImageNet | top-1 accuracy: 75.8 -> 76.6 | Improved |
| use F-norm instead of cosine similarity or L1-norm for tangent distance (ablation) | DCP-NAS-S · ImageNet | top-1 accuracy: 62.5 -> 63.0 | Improved |
| keep pooling operations in search space (ablation) | DCP-NAS-S · ImageNet | top-1 accuracy: 59.1 -> 63.0 | Improved |
| keep skip connections in search space (ablation) | DCP-NAS-S · ImageNet | top-1 accuracy: 60.0 -> 63.0 | Improved |
| keep activations of first depth-wise conv real-valued instead of binarizing (ablation) | DCP-NAS-L · ImageNet | top-1 accuracy: 70.1 -> 72.4 | Improved |
| keep preprocessing activations real-valued instead of binarizing (ablation) | DCP-NAS-S · ImageNet | top-1 accuracy: 60.7 -> 63.0 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 2 on most shared results.
- BATS (Small) — BATS: Binary ArchitecTure Search (ECCV 2020) · better on 4 of 4 shared results
- DARTS 32/32 — DARTS-: Robustly Stepping out of Performance Collapse Without Indicators (ICLR 2021) · better on 0 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
- HEP-NAS: Towards Efficient Few-shot Neural Architecture Search via Hierarchical Edge Partitioning (AAAI 2025 · similarity 0.88)
- DSNAS: Direct Neural Architecture Search Without Parameter Retraining (CVPR 2020 · similarity 0.88)
- Neural Architecture Search on ImageNet in Four GPU Hours: A Theoretically Inspired Perspective (ICLR 2021 · similarity 0.88)
- Generalizing Few-Shot NAS with Gradient Matching (ICLR 2022 · similarity 0.87)
- NAS-FCOS: Fast Neural Architecture Search for Object Detection (CVPR 2020 · similarity 0.87)
- DDPNAS: Efficient Neural Architecture Search via Dynamic Distribution Pruning (IJCV 2023 · similarity 0.87)
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.
