Research map / Architectures and efficiency / Neural architecture search
Towards Improving the Consistency, Efficiency, and Flexibility of Differentiable Neural 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 |
|---|---|---|---|
| Use Engine-cells for search and Transit-cells to carry the derived subgraph (ablation) | Differentiable NAS on CIFAR-10 · CIFAR-10 validation | child-net accuracy (%): 63.97 -> 86.24 | Helped |
| Place the Engine-cell first rather than last (ablation) | EnTranNAS · CIFAR-10 validation | child-net accuracy (%): 83.35 -> 86.24 | Helped |
| Use EnTranNAS or EnTranNAS-DST instead of DARTS | Differentiable NAS · CIFAR-10, six-seed rank correlation | Kendall τ: DARTS -0.47; EnTranNAS 0.33; EnTranNAS-DST 0.60 | Helped |
| Add Engine- and Transit-cells (ablation) | DARTS search · CIFAR-10 search | search cost (GPU-day): 0.73 -> 0.22 | Helped |
| Add feature sharing (ablation) | EnTranNAS with Engine- and Transit-cells · CIFAR-10 search | search cost (GPU-day): 0.22 -> 0.09 | Helped |
| Add bottleneck channel reduction (ablation) | EnTranNAS with feature sharing · CIFAR-10 search | search cost (GPU-day): 0.09 -> 0.06 | Helped |
| Search with 12 operations rather than the standard operation set | EnTranNAS · CIFAR-10 test | test error (%): 2.53$\pm$0.06 -> 2.22$\pm$0.05 | Helped |
| Enable differentiable topology learning with EnTranNAS-DST | EnTranNAS on ImageNet · ImageNet validation | top-1 error (%): 24.3 -> 23.8 | Helped |
| Increase λ from 0.05 to 0.2 (ablation) | EnTranNAS-DST · Pascal VOC 2012 | mIoU (%): 77.1 -> 76.3 | Hurt |
| Increase λ from 0.05 to 0.2 (ablation) | EnTranNAS-DST · ImageNet architecture | edges (normal / reduction): 13 / 14 -> 9 / 8 | Mixed |
Closest papers by meaning
- Improving Differentiable Neural Architecture Search by Encouraging Transferability (ICLR 2023 · similarity 0.90)
- BaLeNAS: Differentiable Architecture Search via the Bayesian Learning Rule (CVPR 2022 · similarity 0.90)
- Neural Architecture Search on ImageNet in Four GPU Hours: A Theoretically Inspired Perspective (ICLR 2021 · similarity 0.89)
- DrNAS: Dirichlet Neural Architecture Search (ICLR 2021 · similarity 0.89)
- iDARTS: Differentiable Architecture Search with Stochastic Implicit Gradients (ICML 2021 · similarity 0.89)
- Unchain the Search Space with Hierarchical Differentiable Architecture Search (AAAI 2021 · similarity 0.89)
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.