Research map / Architectures and efficiency / Neural architecture search
Understanding and Simplifying One-Shot 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 |
|---|---|---|---|
| path dropout with increasing dropout rate schedule during one-shot training (vs. very low or very high constant dropout) (ablation) | one-shot model training · CIFAR-10 one-shot model correlation | one-shot vs stand-alone accuracy correlation | Helped |
| scheduling dropout rate linearly from zero to r=0.1 instead of constant rates | one-shot model training · CIFAR-10 | architecture ranking quality | Helped |
| ghost batch normalization style dropping of different paths per ghost batch | one-shot model training · CIFAR-10 training stability | training stability | Helped |
| searching for smallest architectures above a one-shot accuracy threshold instead of top-accuracy architectures | one-shot architecture selection · CIFAR-10 | parameters vs accuracy: One-Shot Small 96.0 vs One-Shot Top 95.9 at F=64 with fewer parameters (1.3M vs 2.7M at F=32) | Helped |
| one-shot search with smallest-above-threshold selection on ImageNet | one-shot architecture selection · ImageNet | top-1 accuracy vs parameters: One-Shot Small (F=32) 74.2 with 5.1M params vs Random (F=32) 74.1 with 7.7M params | Helped |
| one-shot search with top-ranked architectures vs random architectures | one-shot architecture search · CIFAR-10 | test accuracy: about 0.5% absolute higher (e.g., 96.1 vs 95.8 at F=128) with 1.6x parameters | Helped |
| one-shot search with top architectures vs random architectures | one-shot architecture search · ImageNet | top-1 accuracy: 75.2 vs 74.1 (F=32) | Helped |
| enabling all operations (All On) instead of pruned architectures | one-shot architecture evaluation · CIFAR-10 | test accuracy: All On 96.0 vs One-Shot Top 95.9 at F=64, but ~2x parameters | No clear effect |
| Training the one-shot model for longer (5k to 10k steps) | one-shot model training · CIFAR-10 | one-shot vs stand-alone correlation | No clear effect |
| L2 regularization applied only to parts used by current architecture | one-shot model training · CIFAR-10 | architecture ranking | No clear effect |
Closest papers by meaning
- Building Optimal Neural Architectures using Interpretable Knowledge (CVPR 2024 · similarity 0.89)
- Rethinking Bi-Level Optimization in Neural Architecture Search: A Gibbs Sampling Perspective (AAAI 2021 · similarity 0.88)
- BayesNAS: A Bayesian Approach for Neural Architecture Search (ICML 2019 · similarity 0.88)
- Revisiting Neural Networks for Few-Shot Learning: A Zero-Cost NAS Perspective (ICML 2025 · similarity 0.87)
- Efficient Few-Shot Neural Architecture Search by Counting the Number of Nonlinear Functions (AAAI 2025 · similarity 0.86)
- Prior Knowledge Guided Neural Architecture Generation (ICML 2025 · similarity 0.86)
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.