Research map / Architectures and efficiency / Pruning and efficient tuning
A Main/Subsidiary Network Framework for Simplifying Binary Neural Networks
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 |
|---|---|---|---|
| Use learned main/subsidiary filter selection instead of MSF-Layerwise pruning | Binary NIN · CIFAR-10 | error rate: 19.28% -> 16.89 | Helped |
| Use learned main/subsidiary filter selection instead of MSF-Cascade pruning | Binary VGG-11 · CIFAR-10 | error rate: 18.79% -> 18.03 | Helped |
| Use learned main/subsidiary filter selection instead of MSF-Cascade pruning | Binary ResNet-18 · CIFAR-10 | error rate: 14.63% -> 13.61% | Helped |
| Use learned main/subsidiary filter selection instead of MSF-Cascade pruning | Binary ResNet-18 · ImageNet | error rate: 50.56% -> 49.87% | Helped |
| Prune filters with learned selection rather than retaining the unpruned model | Binary ResNet-18 · ImageNet | error rate: 50.02% -> 49.87% | Helped |
| Use the learned pruning framework rather than XNOR-NET | ResNet-18 · ImageNet | speedup: 10.86$\times$ -> 12.39$\times$ | Helped |
| Use the learned pruning framework rather than XNOR-NET | ResNet-18 · ImageNet | memory usage: 33.70Mbit -> 30.87Mbit | Helped |
| Use a subsidiary-component learning rate of 10^{-3} rather than 10^{-4} (ablation) | Main/subsidiary pruning framework · CIFAR-10 experiments | pruning amount | Mixed |
| Initialize mask PNR at 0.2 rather than 0.4, 0.6, or 1.0 (ablation) | Subsidiary component · Different models and layers | final performance | Hurt |
| Train subsidiary layers bottom-up and retrain the remaining main-network layers after each layer selection (ablation) | Main/subsidiary pruning framework · Binary network pruning | pruning ratio and accuracy balance | Helped |
| Add distillation loss during subsidiary-component training (ablation) | Subsidiary component · Binary network pruning | training stability | Helped |
Closest papers by meaning
- Pruning from Scratch (AAAI 2020 · similarity 0.89)
- RePr: Improved Training of Convolutional Filters (CVPR 2019 · similarity 0.89)
- Provable Filter Pruning for Efficient Neural Networks (ICLR 2020 · similarity 0.89)
- Learning Filter Pruning Criteria for Deep Convolutional Neural Networks Acceleration (CVPR 2020 · similarity 0.89)
- DPFPS: Dynamic and Progressive Filter Pruning for Compressing Convolutional Neural Networks from Scratch (AAAI 2021 · similarity 0.89)
- Entropy Induced Pruning Framework for Convolutional Neural Networks (AAAI 2024 · similarity 0.88)
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.