Research map / Architectures and efficiency / Quantization
Distribution-Sensitive Information Retention for Accurate Binary Neural Network
Haotong Qin, Xiangguo Zhang, Ruihao Gong, Yifu Ding, Yi Xu, Xianglong Liu
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 |
|---|---|---|---|
| add IMB (information maximized binarization) (ablation) | binary ResNet-20 · CIFAR-10 ablation (W1/A1) | accuracy: 83.8 -> 84.9 | Improved |
| add weight standardization to IMB (ablation) | IMB, ResNet-20 · CIFAR-10 ablation (W1/A1) | accuracy: 84.3 -> 84.9 | Improved |
| add shift-based scalars to IMB (ablation) | IMB, ResNet-20 · CIFAR-10 ablation (W1/A1) | accuracy: 84.6 -> 84.9 | Improved |
| add DTE (both stages) (ablation) | binary ResNet-20 · CIFAR-10 ablation (W1/A1) | accuracy: 83.8 -> 85.5 | Improved |
| use only stage 1 of DTE (ablation) | binary ResNet-20 · CIFAR-10 ablation (W1/A1) | accuracy: 83.8 -> 83.6 | Hurt |
| keep at least 10% of weights updatable (eps 10% vs 0%) (ablation) | DTE stage 2 · CIFAR-10 ablation (W1/A1) | accuracy: 84.9 -> 85.1 | Improved |
| add RBD with proposed loss (ablation) | binary ResNet-20 · CIFAR-10 ablation (W1/A1) | accuracy: 83.8 -> 86.2 | Improved |
| distill with MSE or KL instead of proposed RBD loss (ablation) | RBD, ResNet-20 · CIFAR-10 ablation (W1/A1) | accuracy: 86.2 -> 84.9 | Hurt |
| lower eps from 100% to 10% (ablation) | DIR-Net, ResNet-20 · CIFAR-10 epsilon sweep (W1/A1) | accuracy: 84.2 -> 86.8 | Improved |
| lower eps from 10% to 0% (ablation) | DIR-Net, ResNet-20 · CIFAR-10 epsilon sweep (W1/A1) | accuracy: 86.8 -> 86.5 | Hurt |
| combine IMB, DTE and RBD (ablation) | binary ResNet-20 · CIFAR-10 ablation (W1/A1) | accuracy: 83.8 -> 89.0 | Improved |
| use IMB instead of XNOR scaling (ablation) | binary ResNet-20 · CIFAR-10 binarization error | accuracy: 84.8 -> 84.9 | No clear effect |
Closest papers by meaning
- Forward and Backward Information Retention for Accurate Binary Neural Networks (CVPR 2020 · similarity 0.93)
- Network Binarization via Contrastive Learning (ECCV 2022 · similarity 0.90)
- Sparsity-Inducing Binarized Neural Networks (AAAI 2020 · similarity 0.90)
- BiBench: Benchmarking and Analyzing Network Binarization (ICML 2023 · similarity 0.88)
- Training Binary Neural Networks via Gaussian Variational Inference and Low-Rank Semidefinite Programming (NeurIPS 2024 · similarity 0.87)
- Fast and Slow Gradient Approximation for Binary Neural Network Optimization (AAAI 2025 · similarity 0.86)
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.
