Research map / Architectures and efficiency / Quantization
How to Train a Compact Binary Neural Network with High Accuracy?
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 |
|---|---|---|---|
| lower learning rate from 0.01 to 0.0001 (ablation) | BNN · ImageNet | top-5 accuracy: 43.5 -> 61.3 | Helped |
| PReLU instead of ReLU (absorbing scale factors) (ablation) | BNN · ImageNet | top-5 accuracy: 61.3 -> 65.6 | Helped |
| binarize last layer (no scale layer) (ablation) | BNN · ImageNet | top-5 accuracy: 65.6 -> 61.0 | Hurt |
| add scale layer after binarized last layer (ablation) | BNN · ImageNet | top-5 accuracy: 61.0 -> 64.6 | Helped |
| binarize last layer with scale layer (ablation) | BNN · ImageNet | compression rate: 5.98× -> 27.1× | Helped |
| multiple binarizations of activations (m=2) (ablation) | BNN · ImageNet | top-5 accuracy: 64.6 -> 70.4 | Helped |
| expand lower layers of baseline network (ablation) | BNN 2-bit · ImageNet | top-5 accuracy: 70.4 -> 75.6 | Helped |
| bipolar regularizer instead of L2 (ablation) | BNN · ImageNet | training loss stability | Helped |
| full method vs XNOR-net ResNet-18 | binary networks · ImageNet | top-5 accuracy: 73.2 -> 75.6 | Helped |
Closest papers by meaning
- Towards Accurate Binary Convolutional Neural Network (NeurIPS 2017 · similarity 0.89)
- Training binary neural networks with real-to-binary convolutions (ICLR 2020 · similarity 0.88)
- Binarized Neural Networks (NeurIPS 2016 · similarity 0.88)
- Forward and Backward Information Retention for Accurate Binary Neural Networks (CVPR 2020 · similarity 0.88)
- High-Capacity Expert Binary Networks (ICLR 2021 · similarity 0.88)
- Training Binary Neural Networks via Gaussian Variational Inference and Low-Rank Semidefinite Programming (NeurIPS 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.