Research map / Architectures and efficiency / Neural architecture search
DrNAS: Dirichlet Neural 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 |
|---|---|---|---|
| learn a Dirichlet distribution over operation mixing weights instead of directly optimizing point weights (ablation) | differentiable NAS on CIFAR-10 · CIFAR-10 | test error (%): 2.54±0.03 -> 2.46±0.03 | Helped |
| add progressive learning to the Dirichlet architecture search (ablation) | DrNAS · CIFAR-10 | test error (%): 2.54±0.03 -> 2.46±0.03 | Helped |
| add progressive learning to the Dirichlet architecture search (ablation) | DrNAS · ImageNet mobile | top-1 test error (%): 24.2 -> 23.7 | Helped |
| add progressive learning to the Dirichlet architecture search (ablation) | DrNAS · ImageNet mobile | top-5 test error (%): 7.3 -> 7.1 | Helped |
| restrict mixed operations to a partial channel subset rather than using all channels (K=2, 4, or 8) (ablation) | DrNAS partial-channel search · NAS-Bench-201 CIFAR-10 | test accuracy (%): 94.36±0.00 -> 93.49±0.28; 92.85±0.35; 91.06±0.00 | Hurt |
| restrict mixed operations to a partial channel subset rather than using all channels (K=2, 4, or 8) (ablation) | DrNAS partial-channel search · NAS-Bench-201 CIFAR-100 | test accuracy (%): 73.51±0.00 -> 68.48±0.41; 66.68±3.22; 55.11±13.78 | Hurt |
| remove the anchor regularizer (λ=0) (ablation) | DrNAS · NAS-Bench-201 CIFAR-10 | test accuracy (%): 93.78 vs 94.36 | Hurt |
| set anchor regularizer λ to 5e^{-4} (ablation) | DrNAS · NAS-Bench-201 CIFAR-10 | test accuracy (%): 94.01 vs 94.36 | Hurt |
| set anchor regularizer λ to 1e^{-3} (ablation) | DrNAS · NAS-Bench-201 CIFAR-10 | test accuracy (%): 94.36 vs 93.78 | Helped |
| set anchor regularizer λ to 5e^{-3} (ablation) | DrNAS · NAS-Bench-201 CIFAR-10 | test accuracy (%): 94.36 vs 93.78 | Helped |
| set anchor regularizer λ to 1e^{-2} (ablation) | DrNAS · NAS-Bench-201 CIFAR-10 | test accuracy (%): 94.36 vs 93.78 | Helped |
| set anchor regularizer λ to 1e^{-1} (ablation) | DrNAS · NAS-Bench-201 CIFAR-10 | test accuracy (%): 93.76 vs 94.36 | Hurt |
Closest papers by meaning
- BaLeNAS: Differentiable Architecture Search via the Bayesian Learning Rule (CVPR 2022 · similarity 0.91)
- iDARTS: Differentiable Architecture Search with Stochastic Implicit Gradients (ICML 2021 · similarity 0.90)
- DARTS: Differentiable Architecture Search (ICLR 2019 · similarity 0.89)
- Meta Architecture Search (NeurIPS 2019 · similarity 0.89)
- Towards Improving the Consistency, Efficiency, and Flexibility of Differentiable Neural Architecture Search (CVPR 2021 · similarity 0.89)
- Neural Architecture Search With Random Labels (CVPR 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.