Research map / Architectures and efficiency / Quantization
Training wide residual networks for deployment using a single bit for each weight
Mark D. McDonnell
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 |
|---|---|---|---|
| Use 1-bit weights instead of full-precision weights | WRN 20-10 · CIFAR-10, standard augmentation | error rate: 4.22 -> 4.72 | Hurt |
| Use 1-bit weights instead of full-precision weights | WRN 20-10 · CIFAR-100, standard augmentation | error rate: 18.76 -> 19.35 | Hurt |
| Add cutout augmentation (ablation) | 1-bit WRN 20-10 · CIFAR-10 | error rate: 4.72 -> 3.92 | Improved |
| Add cutout augmentation (ablation) | 1-bit WRN 20-10 · CIFAR-100 | error rate: 19.35 -> 18.51 | Improved |
| Do not learn batch-normalization scale and offset parameters (ablation) | Wide ResNets · CIFAR-100 | error rate: around 3% absolute | Improved |
| Do not learn batch-normalization scale and offset parameters (ablation) | Wide ResNets · ImageNet32 | error rate | Hurt |
| Use warm-restart learning-rate schedule (ablation) | 1-bit Wide ResNets · CIFAR-100 | error rate | Improved |
| Use warm-restart learning-rate schedule (ablation) | Full-precision Wide ResNets · CIFAR-100 | error rate | No clear effect |
| Use warm-restart learning-rate schedule (ablation) | Wide ResNets · CIFAR-10/100 | convergence: just 30 epochs | Improved |
| Use 1-bit weights instead of full-precision weights | WRN 20-4 · SVHN | error rate: 1.75 -> 1.93 | Hurt |
| Use 1-bit weights instead of full-precision weights | WRN 20-15 · ImageNet32 | Top-1 error: 38.90 -> 41.26 | Hurt |
| Use 1-bit weights instead of full-precision weights | WRN 18-2.5 · ImageNet, single crop | Top-1 error: 26.92 -> 31.03 | Hurt |
Closest papers by meaning
- Weight Fixing Networks (ECCV 2022 · similarity 0.87)
- Bi-Real Net: Binarizing Deep Network Towards Real-Network Performance (IJCV 2019 · similarity 0.87)
- Efficient Multi-bit Quantization Network Training via Weight Bias Correction and Bit-wise Coreset Sampling (NeurIPS 2025 · similarity 0.86)
- Explicit Loss-Error-Aware Quantization for Low-Bit Deep Neural Networks (CVPR 2018 · similarity 0.86)
- OvSW: Overcoming Silent Weights for Accurate Binary Neural Networks (ECCV 2024 · similarity 0.86)
- Bi-Real Net: Enhancing the Performance of 1-bit CNNs with Improved Representational Capability and Advanced Training Algorithm (ECCV 2018 · 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.
