Research map / Architectures and efficiency / Quantization
Sparsity-Inducing Binarized Neural Networks
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 |
|---|---|---|---|
| quantize activations to {0,+1} (sparsity-inducing) instead of sign {-1,+1}, with trainable thresholds | binarized AlexNet · ImageNet val | Top-1 accuracy (%): 44.2 -> 50.5 | Helped |
| quantize activations to {0,+1} in ResNet-18 binarization | binarized ResNet-18 · ImageNet val | Top-1 accuracy (%): 51.2 -> 58.9 | Helped |
| make binarization threshold trainable instead of fixed hand-crafted (ablation) | Si-BNN AlexNet · ImageNet val | Top-1 accuracy (%): 49.2 -> 50.5 | Helped |
| set backward asymmetry rho to 0.1 (vs 0.3) (ablation) | Si-BNN AlexNet · ImageNet val | Top-1 accuracy (%): 50.5 -> 41.9 | Hurt |
| set rho to 1 (symmetric clip) vs 0.3 (ablation) | Si-BNN AlexNet · ImageNet val | Top-1 accuracy (%): 50.5 -> 49.4 | Hurt |
| raise hand-set sparsity threshold from <0.4 up to 0.8 (ablation) | Si-BNN AlexNet · ImageNet val | Top-1 accuracy (%): 45.8 at threshold 0.8 | Mixed |
| Si-BNN (1-bit act) vs HWGQ (2-bit act) | ResNet-18 · ImageNet val | Top-1 accuracy (%): 59.6 -> 59.7 | No clear effect |
| Si-BNN instead of BNN on binarized MLP | MNIST · MNIST | error rate (%): 1.40 -> 1.26 | Helped |
| Si-BNN instead of BNN on CIFAR-10 | BNN architecture · CIFAR-10 | error rate (%): 10.15 -> 9.80 | Helped |
Closest papers by meaning
- Understanding Neural Network Binarization with Forward and Backward Proximal Quantizers (NeurIPS 2023 · similarity 0.90)
- Binarized Neural Networks (NeurIPS 2016 · similarity 0.89)
- Forward and Backward Information Retention for Accurate Binary Neural Networks (CVPR 2020 · similarity 0.88)
- BD-Net: Has Depth-Wise Convolution Ever Been Applied in Binary Neural Networks? (AAAI 2026 · similarity 0.88)
- BiPer: Binary Neural Networks using a Periodic Function (CVPR 2024 · similarity 0.87)
- 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.