Research map / Architectures and efficiency / Neural architecture search
Progressive Neural Architecture Search
Chenxi Liu, Barret Zoph, Maxim Neumann, Jonathon Shlens, Wei Hua, Li-Jia Li, Li Fei-Fei, Alan Yuille, Jonathan Huang, Kevin Murphy
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (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 |
|---|---|---|---|
| progressive search with learned predictor instead of NAS RL (ablation) | PNAS · CIFAR-10 val (top 1) | models evaluated: 5808 -> 1160 | Improved |
| progressive search instead of NAS, total example cost | PNAS · CIFAR-10 | compute: about 8 times faster; 21 times less compute | Improved |
| progressive search instead of random search (ablation) | PNAS · CIFAR-10 val | top-M accuracy | Improved |
| use MLP-ensemble instead of RNN-ensemble predictor (ablation) | PNAS · CIFAR-10 val | search efficiency: speedup 1.9 -> 5.0 | Improved |
| add more blocks to cells (b=1 to b=5) (ablation) | PNASNet · CIFAR-10 test | top-1 error: 4.01 -> 3.41 | Improved |
| search beyond 5 blocks up to B=10 (ablation) | PNAS · CIFAR-10 val | top-M accuracy | Improved |
| compare PNASNet-5 with NASNet-A | PNASNet-5 · ImageNet (Mobile) | top-1 accuracy: 74.0 -> 74.2 | Improved |
| compare PNASNet-5 with NASNet-A | PNASNet-5 · ImageNet (Large) | top-1 accuracy: 82.7 -> 82.9 | Improved |
| use PNASNet-5 instead of AmoebaNet-C | PNASNet-5 · ImageNet (Mobile) | top-1 accuracy: 75.7 -> 74.2 | Hurt |
| PNAS cell vs NASNet-A on CIFAR-10 | PNASNet-5 · CIFAR-10 test | top-1 error: 3.41 vs 3.41 | No clear effect |
| CIFAR-10 accuracy as proxy for ImageNet accuracy | Architecture search · CIFAR-10 to ImageNet | rank correlation: rho 0.727 | Improved |
| ensemble predictors for extrapolation to larger cells | MLP-ensemble · Predictor extrapolation | Spearman rank correlation: 0.164 vs 0.113 at b=1; 0.468 vs 0.444 at b=4 | Mixed |
Closest papers by meaning
- Hierarchical Representations for Efficient Architecture Search (ICLR 2018 · similarity 0.89)
- Neural Architecture Optimization (NeurIPS 2018 · similarity 0.88)
- Evolving Search Space for Neural Architecture Search (ICCV 2021 · similarity 0.88)
- SGAS: Sequential Greedy Architecture Search (CVPR 2020 · similarity 0.88)
- Designing Neural Network Architectures using Reinforcement Learning (ICLR 2017 · similarity 0.87)
- Boosting Order-Preserving and Transferability for Neural Architecture Search: a Joint Architecture Refined Search and Fine-tuning Approach (CVPR 2024 · similarity 0.87)
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.
