Research map / Architectures and efficiency / Neural architecture search
Unchain the Search Space with Hierarchical Differentiable 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 |
|---|---|---|---|
| Search stage-specific cell structures instead of one shared cell (ablation) | DARTS · CIFAR10 | test error (%): 2.76 -> 2.66 | Helped |
| Search stage-level connections in addition to the shared cell (ablation) | DARTS · CIFAR10 | test error (%): 2.76 -> 2.58 | Helped |
| Use stage-specific macro-architecture search rather than shared stage structure (ablation) | H^s-DAS · CIFAR10 | test error (%): 2.58 -> 2.30 | Helped |
| Use stage-specific cell search together with stage-specific macro-architecture search (ablation) | H^s-DAS · ImageNet | top-1 test error (%): 24.50 -> 25.80 | Hurt |
| Constrain cell predecessors with a sliding window, reducing m from 7 to 3 (ablation) | H^s-DAS · CIFAR10 | test error (%): 3.08 -> 2.58 | Helped |
| Add a depth loss to encourage deeper stage structures (ablation) | H^s-DAS · CIFAR10 | test error (%): 2.59 -> 2.34 | Helped |
| Increase depth-loss weight from δ=1 to δ=1.5 (ablation) | H^s-DAS · CIFAR10 | test error (%): 2.34 -> 2.53 | Hurt |
| Change stage-level operations while keeping the connections fixed (ablation) | H^s-DAS · CIFAR10 | top-1 accuracy: 97.49 -> 97.48 | No clear effect |
| Replace the searched cell with a P-DARTS cell (ablation) | H^s-DAS · CIFAR10 | test error (%): 2.41 -> 2.30 | Helped |
| Replace the searched cell with a NoisyDARTS cell (ablation) | H^s-DAS · CIFAR10 | test error (%): 2.41 -> 2.34 | Helped |
| Combine H^s-DAS with AutoAugment | H^s-DAS · CIFAR10 | test error (%): 2.41 -> 1.99 | Helped |
Closest papers by meaning
- Towards Improving the Consistency, Efficiency, and Flexibility of Differentiable Neural Architecture Search (CVPR 2021 · similarity 0.89)
- DOTS: Decoupling Operation and Topology in Differentiable Architecture Search (CVPR 2021 · similarity 0.89)
- On Redundancy and Diversity in Cell-based Neural Architecture Search (ICLR 2022 · similarity 0.88)
- PC-DARTS: Partial Channel Connections for Memory-Efficient Architecture Search (ICLR 2020 · similarity 0.88)
- Towards Neural Architecture Search through Hierarchical Generative Modeling (ICML 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.