Research map / Architectures and efficiency / Neural architecture search
Path-Level Network Transformation for Efficient Architecture Search
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 DropPath and train for 600 epochs (ablation) | TreeCell-A + PyramidNet (N=18, α=84, G=2) · CIFAR-10 test | test error (%): 3.14 -> 2.99 | Helped |
| Add Cutout alongside DropPath (ablation) | TreeCell-A + PyramidNet (N=18, α=84, G=2) · CIFAR-10 test | test error (%): 2.99 -> 2.49 | Helped |
| Increase PyramidNet α from 84 to 150 (ablation) | TreeCell-A + PyramidNet with DropPath and Cutout · CIFAR-10 test | test error (%): 2.49 -> 2.30 | Helped |
| Increase DenseNet depth from N=6 to N=16 (ablation) | TreeCell-A + DenseNet (k=48, G=2) · CIFAR-10 test | test error (%): 3.64 -> 3.35 | Helped |
| Use TreeCell-A instead of TreeCell-B in PyramidNet (ablation) | PyramidNet (N=18, α=84, G=2) · CIFAR-10 test | test error (%): 3.40 -> 3.14 | Helped |
| Use TreeCell-A instead of TreeCell-B in DenseNet (ablation) | DenseNet (N=6, k=48, G=2) · CIFAR-10 test | test error (%): 3.71 -> 3.64 | Helped |
| Search using RL instead of random search | Path-Level Efficient Architecture Search · CIFAR-10 validation, starting from the DenseNet base | validation accuracy: 0.16% | Helped |
| Use the best RL-discovered cell instead of the best random-search cell | CIFAR-10 cell search from the DenseNet base · CIFAR-10 test | test error (%): 3.98 -> 3.64 | Helped |
| Transfer TreeCell-A to CondenseNet | CondenseNet (G1=4, G3=8) · ImageNet Mobile | top-1 error (%): 26.2 -> 25.5 | Helped |
| Search from an identity chain rather than reuse a successful base architecture (ablation) | Path-Level Efficient Architecture Search · CIFAR-10 test | test error (%): 3.14 -> 3.93 | Hurt |
| Use the RL-discovered cell instead of the best random-search cell from scratch | Identity-chain cell search · CIFAR-10 test | test error (%): 4.44 -> 3.93 | Helped |
Closest papers by meaning
- Efficient Architecture Search by Network Transformation (AAAI 2018 · similarity 0.91)
- Cream of the Crop: Distilling Prioritized Paths For One-Shot Neural Architecture Search (NeurIPS 2020 · similarity 0.88)
- Hierarchical Representations for Efficient Architecture Search (ICLR 2018 · similarity 0.87)
- Meta Architecture Search (NeurIPS 2019 · similarity 0.87)
- Boosting Order-Preserving and Transferability for Neural Architecture Search: a Joint Architecture Refined Search and Fine-tuning Approach (CVPR 2024 · similarity 0.86)
- HEP-NAS: Towards Efficient Few-shot Neural Architecture Search via Hierarchical Edge Partitioning (AAAI 2025 · 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.