Research map / Architectures and efficiency / Neural architecture search
FP-NAS: Fast Probabilistic 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 |
|---|---|---|---|
| adapt architecture sample count to distribution entropy | PARSEC fixed sampling · FBNetV2-F, ImageNet-100 search | search time: 60% fewer architectures; 1.8× faster | Helped |
| adapt architecture sample count to distribution entropy | PARSEC fixed sampling · FBNetV2-F-Fine, ImageNet-1K validation | top-1 accuracy: 66.3 -> 67.2; +0.9% | Helped |
| increase adaptive sampling λ from 1/4 to 1/2 (ablation) | FP-NAS adaptive sampling · FBNetV2-F search | accuracy-to-complexity trade-off | No clear effect |
| start with a factorized distribution, then convert to a joint distribution | FP-NAS · FBNetV2-F, ImageNet-1K validation | search cost: 2.4 -> 2.1 GPU days; 1.2× faster | Helped |
| use a factorized distribution throughout instead of switching to joint (ablation) | FP-NAS · FBNetV2-F, ImageNet-1K validation | top-1 accuracy: 68.6 -> 67.6 (-1.0) | Hurt |
| convert factorized distribution to joint at epoch 150 rather than 80 (ablation) | FP-NAS mixed distribution · FBNetV2-F, ImageNet-1K validation | top-1 accuracy: 68.6 -> 68.2 (-0.4) | Hurt |
| replace the searched architecture with FBNetV2-F1 | FP-NAS-S1 · ImageNet-1K validation, input 128 | top-1 accuracy: 68.3 -> 68.6 (+0.3) in favor of FP-NAS-S1 | Hurt |
| replace the searched architecture with FBNetV2-F2 | FP-NAS-S2 · ImageNet-1K validation, input 160 | top-1 accuracy: 71.1 -> 71.3 (+0.2) in favor of FP-NAS-S2 | Hurt |
| search SA placement and split count instead of applying SA uniformly | FP-NAS-S models · ImageNet-1K validation, input 128 | top-1 accuracy: 69.3 -> 70.0; 69.5 -> 70.0 | Helped |
| uniformly replace SE with SA using 2 splits (ablation) | FP-NAS-S1 · ImageNet-1K validation, input 128 | top-1 accuracy: 68.6 -> 69.3 | Helped |
| uniformly replace SE with SA using 4 splits (ablation) | FP-NAS-S1 · ImageNet-1K validation, input 128 | top-1 accuracy: 68.6 -> 69.5 | Helped |
| train FP-NAS-L2 with simple distillation (ablation) | FP-NAS-L2 · ImageNet-1K validation | top-1 accuracy: 80.7 -> 81.6 | Helped |
Closest papers by meaning
- Neural Architecture Search on ImageNet in Four GPU Hours: A Theoretically Inspired Perspective (ICLR 2021 · similarity 0.91)
- Posterior-Guided Neural Architecture Search (AAAI 2020 · similarity 0.91)
- NAS-Bench-201: Extending the Scope of Reproducible Neural Architecture Search (ICLR 2020 · similarity 0.91)
- FBNetV3: Joint Architecture-Recipe Search Using Predictor Pretraining (CVPR 2021 · similarity 0.91)
- BaLeNAS: Differentiable Architecture Search via the Bayesian Learning Rule (CVPR 2022 · similarity 0.90)
- Efficient Few-Shot Neural Architecture Search by Counting the Number of Nonlinear Functions (AAAI 2025 · similarity 0.90)
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.