Research map / Architectures and efficiency / Neural architecture search
b-DARTS: Beta-Decay Regularization for 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 Beta-Decay regularization to DARTS | DARTS on NAS-Bench-201 · CIFAR-10, CIFAR-100, and ImageNet16-120 test | accuracy: 93.80±0.40 -> 94.36±0.00; 71.53±1.51 -> 73.51±0.00; 45.12±0.82 -> 46.34±0.00 | Helped |
| Search on CIFAR-10 using Beta-Decay regularization, then transfer the architecture | DARTS on NAS-Bench-201 · CIFAR-100 test | accuracy: 83.76±0.22 | Helped |
| Use Beta-Global loss instead of the local Beta-Decay loss (ablation) | β-DARTS · NAS-Bench-201 CIFAR-10 test | accuracy: 93.83/94.36 | No clear effect |
| Use Beta-Zero loss instead of Beta-Decay (ablation) | β-DARTS · NAS-Bench-201 CIFAR-100 test | accuracy: 73.51/71.88 | Hurt |
| Increase the regularization coefficient linearly (ablation) | β-DARTS · CIFAR-10 | accuracy: 93.83/94.36/94.36/94.36 | Helped |
| Use constant rather than linearly increased regularization weights (ablation) | β-DARTS · CIFAR-10 | accuracy: 88.02/93.31/94.36/93.31 | Mixed |
| Use linearly decreasing rather than linearly increased regularization weights (ablation) | β-DARTS · CIFAR-10 | accuracy: 93.31/90.65/76.88/54.30 | Hurt |
| Regularize architecture parameters α with L2 or weight decay instead of regularizing activated parameters β (ablation) | DARTS · DARTS search | search robustness | Hurt |
Closest papers by meaning
- Stabilizing Differentiable Architecture Search via Perturbation-based Regularization (ICML 2020 · similarity 0.92)
- $\Lambda$-DARTS: Mitigating Performance Collapse by Harmonizing Operation Selection among Cells (ICLR 2023 · similarity 0.92)
- iDARTS: Differentiable Architecture Search with Stochastic Implicit Gradients (ICML 2021 · similarity 0.92)
- Operation-Level Early Stopping for Robustifying Differentiable NAS (NeurIPS 2023 · similarity 0.90)
- EG-NAS: Neural Architecture Search with Fast Evolutionary Exploration (AAAI 2024 · similarity 0.90)
- IS-DARTS: Stabilizing DARTS through Precise Measurement on Candidate Importance (AAAI 2024 · 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.