Research map / Architectures and efficiency / Neural architecture search
Fast and Practical Neural Architecture Search
Jiequan Cui, Pengguang Chen, Ruiyu Li, Shu Liu, Xiaoyong Shen, Jiaya Jia
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 |
|---|---|---|---|
| increase E (candidates trained independently) from 5 to 10 (ablation) | FPNAS · CIFAR-10 search hyperparameters | top-1 accuracy (%): 95.38 -> 96.01 | Improved |
| increase E from 10 to 20 (ablation) | FPNAS · CIFAR-10 search hyperparameters | top-1 accuracy (%): 96.01 -> 96.01 | No clear effect |
| increase Q (independent training epochs) from 2 to 4 (ablation) | FPNAS · CIFAR-10 search hyperparameters | top-1 accuracy (%): 95.80 -> 96.01 | Improved |
| increase Q from 4 to 8 (ablation) | FPNAS · CIFAR-10 search hyperparameters | top-1 accuracy (%): 96.01 -> 95.93 | No clear effect |
| increase P (shared-weight epochs) from 2 to 4 (ablation) | FPNAS · CIFAR-10 search hyperparameters | top-1 accuracy (%): 95.59 -> 96.01 | Improved |
| increase P from 4 to 8 (ablation) | FPNAS · CIFAR-10 search hyperparameters | top-1 accuracy (%): 96.01 -> 95.48 | Hurt |
| use different block topologies at different positions instead of one repeated block (ablation) | simple 5-block CIFAR-10 networks · CIFAR-10 (27 small networks) | accuracy (%) | Improved |
| use fast alternating block search instead of reinforcement learning (ablation) | FPNAS search · CIFAR-10 (40 epochs training) | top-1 and top-5 accuracy | No clear effect |
| search networks directly with FPNAS instead of using handcrafted MobileNet V2 (1x) | FPNASNet · ImageNet val | top-1 accuracy (%): 72.00 -> 73.34 | Improved |
| use FPNASNet encoder instead of MobileNet V2 in PSPNet | FPNASNet · ADE20K val (single-scale testing) | mean IoU (%): 35.75 -> 36.76 | Improved |
| add multiscale testing (ablation) | FPNASNet · ADE20K val | mean IoU (%): 36.76 -> 37.40 | Improved |
| search with FPNAS instead of MnasNet for ~300M FLOPs models | FPNASNet · ImageNet val | top-1 accuracy (%): 74.00 -> 73.34 | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- ShuffleNet V2 — ShuffleNet V2: Practical Guidelines for Efficient CNN Architecture Design (ECCV 2018) · better on 2 of 2 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Neural Architecture Search on ImageNet in Four GPU Hours: A Theoretically Inspired Perspective (ICLR 2021 · similarity 0.91)
- DDPNAS: Efficient Neural Architecture Search via Dynamic Distribution Pruning (IJCV 2023 · similarity 0.90)
- ISTA-NAS: Efficient and Consistent Neural Architecture Search by Sparse Coding (NeurIPS 2020 · similarity 0.90)
- AtomNAS: Fine-Grained End-to-End Neural Architecture Search (ICLR 2020 · similarity 0.90)
- MathNAS: If Blocks Have a Role in Mathematical Architecture Design (NeurIPS 2023 · similarity 0.90)
- DA-NAS: Data Adapted Pruning for Efficient Neural Architecture Search (ECCV 2020 · 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.
