Research map / Architectures and efficiency / Neural architecture search
Practical Block-Wise Neural Network Architecture Generation
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 |
|---|---|---|---|
| Search reusable network blocks and stack them, rather than searching the whole network layer by layer | BlockQNN · CIFAR-10 | error rate: 3.54 vs 3.46 | Mixed |
| Transfer the CIFAR-learned block to ImageNet with little modification | BlockQNN · ImageNet-1K | Top-1 error: 23.0 -> 22.6 | Helped |
| Use a limited-parameter block and increase the number of filters (ablation) | Block-QNN-S · CIFAR-10 | error rate: 4.38 -> 3.54 | Helped |
| Combine ReLU, convolution, and batch normalization as a Pre-activation Convolutional Cell instead of searching the components separately (ablation) | BlockQNN · CIFAR-100 block search | early-stop accuracy | Helped |
| Use shaped intermediate rewards rather than setting intermediate rewards to zero (ablation) | BlockQNN Q-learning · Block search | convergence speed | Helped |
| Use FLOPs and graph density penalties with early-stop accuracy as the search reward (ablation) | BlockQNN · Block search | reward correlation with final accuracy | Helped |
| Use Q-learning instead of random search | BlockQNN · CIFAR-100 | performance of best generated model: over 1% | Helped |
| Increase filters in the limited-parameter network (ablation) | Block-QNN-S · CIFAR-100 | error rate: 20.65 -> 18.06 | Helped |
Closest papers by meaning
- Designing Neural Network Architectures using Reinforcement Learning (ICLR 2017 · similarity 0.86)
- BlockSwap: Fisher-guided Block Substitution for Network Compression on a Budget (ICLR 2020 · similarity 0.85)
- Blockout: Dynamic Model Selection for Hierarchical Deep Networks (CVPR 2016 · similarity 0.85)
- Diverse Branch Block: Building a Convolution as an Inception-Like Unit (CVPR 2021 · similarity 0.84)
- Efficient ConvBN Blocks for Transfer Learning and Beyond (ICLR 2024 · similarity 0.83)
- AsCAN: Asymmetric Convolution-Attention Networks for Efficient Recognition and Generation (NeurIPS 2024 · similarity 0.83)
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.