Research map / Architectures and efficiency / Quantization
ReCU: Reviving the Dead Weights in Binary Neural Networks
Zihan Xu, Mingbao Lin, Jianzhuang Liu, Jie Chen, Ling Shao, Yue Gao, Yonghong Tian, Rongrong Ji
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 |
|---|---|---|---|
| apply ReCU clamp with fixed tau=0.92 instead of no clamping (tau=1.00) (ablation) | binary ResNet-18 · CIFAR-100, binary ResNet-18 | top-1 accuracy: 67.55 -> 68.47 | Improved |
| lower fixed tau to 0.82, the quantization-error minimum (ablation) | binary ResNet-18 · CIFAR-100, binary ResNet-18 | top-1 accuracy: 68.47 -> 66.50 | Hurt |
| lower fixed tau to 0.78 (ablation) | binary ResNet-18 · CIFAR-100, binary ResNet-18 | top-1 accuracy: 68.47 -> 59.09 | Hurt |
| replace fixed tau=0.92 with exponential scheduler 0.85/0.99 (ablation) | ReCU · CIFAR-100, binary ResNet-18 | top-1 accuracy: 68.47 -> 68.69 | Improved |
| use end value 0.99 instead of 1.00 in scheduler (start 0.85) (ablation) | ReCU scheduler · CIFAR-100, binary ResNet-18 | top-1 accuracy: 68.55 -> 68.69 | Improved |
| add generalized weight standardization with b*=sqrt(2)/2 (ablation) | ReCU · CIFAR-100, binary ResNet-18 | top-1 accuracy: 66.13 -> 68.69 | Improved |
| raise b* from 0.2 to 2 (ablation) | ReCU standardization · CIFAR-100, binary ResNet-18 | top-1 accuracy: 68.10 -> 69.02 | Improved |
| raise b* from 2 to 5 (ablation) | ReCU standardization · CIFAR-100, binary ResNet-18 | top-1 accuracy: 69.02 -> 68.48 | Hurt |
| use ReCU vs vanilla BNN at 100 training epochs | binary ResNet-18 · CIFAR-100, 100 epochs | top-1 accuracy: 52.1 -> 66.3 | Improved |
| use ReCU vs vanilla BNN at 600 training epochs | binary ResNet-18 · CIFAR-100, 600 epochs | top-1 accuracy: 62.0 -> 69.1 | Improved |
| use ReCU instead of RBNN | binary ResNet-18 · ImageNet | top-1 accuracy: 59.9 -> 61.0 | Improved |
| use ReCU instead of RBNN | binary ResNet-34 · ImageNet | top-1 accuracy: 63.1 -> 65.1 | Improved |
Later papers that compared with this method
8 later accepted papers report ReCU as a baseline in their results tables; 8 of them report a better number on most of the results they share with it.
- BHViT: Binarized Hybrid Vision Transformer (CVPR 2025 · better on 1 of 1 shared result)
- BiPer: Binary Neural Networks using a Periodic Function (CVPR 2024 · better on 5 of 6 shared results)
- OvSW: Overcoming Silent Weights for Accurate Binary Neural Networks (ECCV 2024 · better on 5 of 5 shared results)
- MST-compression: Compressing and Accelerating Binary Neural Networks with Minimum Spanning Tree (ICCV 2023 · better on 4 of 4 shared results)
- Resilient Binary Neural Network (AAAI 2023 · better on 3 of 3 shared results)
- INSTA-BNN: Binary Neural Network with INSTAnce-aware Threshold (ICCV 2023 · better on 1 of 1 shared result)
- AdaBin: Improving Binary Neural Networks with Adaptive Binary Sets (ECCV 2022 · better on 3 of 3 shared results)
- BiMLP: Compact Binary Architectures for Vision Multi-Layer Perceptrons (NeurIPS 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
- OvSW: Overcoming Silent Weights for Accurate Binary Neural Networks (ECCV 2024 · similarity 0.89)
- Resilient Binary Neural Network (AAAI 2023 · similarity 0.89)
- Latent Weights Do Not Exist: Rethinking Binarized Neural Network Optimization (NeurIPS 2019 · similarity 0.88)
- SA-BNN: State-Aware Binary Neural Network (AAAI 2021 · similarity 0.88)
- 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)
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.
