Research map / Architectures and efficiency / Neural architecture search
DOTS: Decoupling Operation and Topology in Differentiable Architecture Search
Check 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 explicit topology search to DARTS (ablation) | DARTS · CIFAR100 | test accuracy (%): 80.74 -> 83.07 | Helped |
| Add topology search to gradient-based NAS (ablation) | DARTS · CIFAR10 | test accuracy (%): 97.02±0.12 -> 97.40±0.09 | Helped |
| Use topology search with the existing method (ablation) | GDAS · CIFAR100 | test accuracy (%): 82.75 -> 83.01 | Helped |
| Replace PC-DARTS edge weights with edge-combination topology weights (ablation) | DOTS · CIFAR100, 2-edge constraint | test accuracy (%): 82.98 -> 83.72 | Helped |
| Remove the fixed edge-number constraint (ablation) | DOTS · CIFAR100 | test accuracy (%): 83.72 -> 83.92 | Helped |
| Use edge-combination weights instead of edge-level sigmoid weights (ablation) | DOTS · CIFAR100, arbitrary edges | test accuracy (%): 81.02 -> 83.92 | Helped |
| Preserve topology-related operations alongside topology-agnostic operations (ablation) | DOTS operation search · CIFAR100 | test accuracy (%): 83.07 -> 83.72 | Helped |
| Retain two strongest operations per edge without grouping (ablation) | DOTS operation search · CIFAR100 | test accuracy (%): 83.07 -> 82.96 | No clear effect |
| Add more groups among topology-agnostic operations (ablation) | DOTS operation search · CIFAR100 | test accuracy (%): 83.72 -> 83.55 | No clear effect |
| Search with edge-combination weights rather than operation weights for topology ranking | DOTS · CIFAR10 and CIFAR100 | Kendall Tau: Tau=0.73 and Tau=0.71 | Helped |
| Search the architecture directly on ImageNet rather than transferring a CIFAR10-searched cell | DOTS · ImageNet | top-1 accuracy (%): 75.7 -> 76.0 | Helped |
Closest papers by meaning
- Unchain the Search Space with Hierarchical Differentiable Architecture Search (AAAI 2021 · similarity 0.89)
- DiNTS: Differentiable Neural Network Topology Search for 3D Medical Image Segmentation (CVPR 2021 · similarity 0.86)
- $\Lambda$-DARTS: Mitigating Performance Collapse by Harmonizing Operation Selection among Cells (ICLR 2023 · similarity 0.86)
- On Redundancy and Diversity in Cell-based Neural Architecture Search (ICLR 2022 · similarity 0.85)
- PC-DARTS: Partial Channel Connections for Memory-Efficient Architecture Search (ICLR 2020 · similarity 0.85)
- EG-NAS: Neural Architecture Search with Fast Evolutionary Exploration (AAAI 2024 · similarity 0.84)
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.