Research map / Architectures and efficiency / Quantization
OvSW: Overcoming Silent Weights for Accurate Binary Neural Networks
jingyang xiang, Zuohui Chen, Siqi Li, Qing Wu, Yong 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 Adaptive Gradient Scaling (AGS) (ablation) | binarized ResNet18 · CIFAR100 | top-1 accuracy (%): 65.23 -> 69.61 | Improved |
| add Silence Awareness Decaying (SAD) (ablation) | binarized ResNet18 · CIFAR100 | top-1 accuracy (%): 65.23 -> 69.45 | Improved |
| add SAD on top of AGS (ablation) | OvSW · CIFAR100 | top-1 accuracy (%): 69.61 -> 69.77 | Improved |
| add OvSW as plug-and-play module to AdaBin | AdaBin · CIFAR100 | top-1 accuracy (%): 70.56 -> 72.56 | Improved |
| add OvSW as plug-and-play module to RBNN | RBNN · CIFAR100 | top-1 accuracy (%): 67.15 -> 69.98 | Improved |
| use OvSW instead of vanilla training for fast convergence | binarized ResNet18 · CIFAR100 (60 epochs) | top-1 accuracy (%): 55.28 -> 66.37 | Improved |
| set AGS threshold lambda too small or too large (best 0.04) (ablation) | AGS · CIFAR100 | top-1 accuracy | Mixed |
| apply OvSW to ResNet18 vs ReCU | OvSW · ImageNet1K | top-1 accuracy (%): 61.0 -> 61.6 | Improved |
| reduce silent weights fraction with OvSW (layer4.conv2) | binarized ResNet18 · CIFAR100 weight flipping | ratio of never-flipped weights: 54.07% -> 2.03% | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- ReCU — ReCU: Reviving the Dead Weights in Binary Neural Networks (ICCV 2021) · better on 5 of 5 shared results
- AdaBin — AdaBin: Improving Binary Neural Networks with Adaptive Binary Sets (ECCV 2022) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- ReCU: Reviving the Dead Weights in Binary Neural Networks (ICCV 2021 · similarity 0.89)
- SiMaN: Sign-to-Magnitude Network Binarization (TPAMI 2022 · similarity 0.88)
- SA-BNN: State-Aware Binary Neural Network (AAAI 2021 · similarity 0.88)
- Latent Weights Do Not Exist: Rethinking Binarized Neural Network Optimization (NeurIPS 2019 · similarity 0.88)
- Understanding weight-magnitude hyperparameters in training binary networks (ICLR 2023 · similarity 0.88)
- Training Binary Neural Networks via Gaussian Variational Inference and Low-Rank Semidefinite Programming (NeurIPS 2024 · similarity 0.85)
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.
