Research map / Architectures and efficiency / Quantization
SA-BNN: State-Aware Binary Neural Network
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 |
|---|---|---|---|
| Add state-aware gradients to the baseline (ablation) | Binary ResNet-18 · ImageNet | Top-1 accuracy: 53.4 -> 60.6 | Helped |
| Add state-aware gradients to the baseline (ablation) | Binary ResNet-34 · ImageNet | Top-1 accuracy: 58.5 -> 63.1 | Helped |
| Add state-aware gradients to the baseline (ablation) | Binary ResNet-50 · ImageNet | Top-1 accuracy: 54.8 -> 67.5 | Helped |
| Add pre-binarization batch normalization to the state-aware model (ablation) | Binary ResNet-18 · ImageNet | Top-1 accuracy: 60.6 -> 61.3 | Helped |
| Add pre-binarization batch normalization to the state-aware model (ablation) | Binary ResNet-34 · ImageNet | Top-1 accuracy: 63.1 -> 65.0 | Helped |
| Add pre-binarization batch normalization to the state-aware model (ablation) | Binary ResNet-50 · ImageNet | Top-1 accuracy: 67.5 -> 68.5 | Helped |
| Add additional shortcuts to the state-aware model with pre-binarization normalization (ablation) | Binary ResNet-18 · ImageNet | Top-1 accuracy: 61.3 -> 61.7 | Helped |
| Add additional shortcuts to the state-aware model with pre-binarization normalization (ablation) | Binary ResNet-34 · ImageNet | Top-1 accuracy: 65.0 -> 65.5 | Helped |
| Add additional shortcuts to the state-aware model with pre-binarization normalization (ablation) | Binary ResNet-50 · ImageNet | Top-1 accuracy: 68.5 -> 68.7 | Helped |
| Use distinct learnable gradient coefficients for the two binary states rather than equal coefficients (ablation) | SA-BNN ResNet-18 · ImageNet | Top-1 accuracy | Helped |
| Initialize both learnable state coefficients to 0.40 rather than fixing τ1=1 (ablation) | SA-BNN ResNet-18 · ImageNet | Top-1 accuracy: 60.8 | Helped |
| Use state-aware gradients instead of state-consistent gradients (ablation) | Binary ResNet-18 · ImageNet | sequential weight-flip ratio | Helped |
Closest papers by meaning
- Resilient Binary Neural Network (AAAI 2023 · similarity 0.89)
- Binarized Neural Networks (NeurIPS 2016 · similarity 0.87)
- Training Binary Neural Networks via Gaussian Variational Inference and Low-Rank Semidefinite Programming (NeurIPS 2024 · similarity 0.86)
- A&B BNN: Add&Bit-Operation-Only Hardware-Friendly Binary Neural Network (CVPR 2024 · similarity 0.86)
- Fast and Slow Gradient Approximation for Binary Neural Network Optimization (AAAI 2025 · similarity 0.84)
- BiPer: Binary Neural Networks using a Periodic Function (CVPR 2024 · similarity 0.84)
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.