Research map / Architectures and efficiency / Quantization
Learnable Companding Quantization for Accurate Low-Bit Neural Networks
Check 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 |
|---|---|---|---|
| Learn quantization levels with a companding function instead of fixed conventional quantization | LCQ · ImageNet validation, ResNet-50 W2/A2 | top-1 accuracy (%): 73.7 -> 75.1 | Helped |
| Learn quantization levels with a companding function instead of fixed conventional quantization | LCQ · ImageNet validation, ResNet-50 W4/A4 | top-1 accuracy (%): 76.7 -> 76.6 | Hurt |
| Add limited weight normalization (LWN) (ablation) | LCQ, ResNet-20 · CIFAR-10 W2/A2 | top-1 accuracy (%): 91.4 -> 91.8 | Helped |
| Add limited weight normalization (LWN) (ablation) | LCQ, ResNet-20 · CIFAR-10 W3/A3 | top-1 accuracy (%): 92.6 -> 92.7 | Helped |
| Add limited weight normalization (LWN) (ablation) | LCQ, ResNet-18 · ImageNet W4/A4 | top-1 accuracy (%): 71.5 -> 71.5 | No clear effect |
| Increase the companding-function interval count from 4 to 16 (ablation) | LCQ, ResNet-20 · CIFAR-100 | accuracy | Helped |
| Reduce outer bit-widths from 8/8 to 6/6 (ablation) | LCQ W3/A3, ResNet-20 · CIFAR-10 | top-1 accuracy (%): 92.8 -> 92.8 | No clear effect |
| Reduce outer bit-widths from 8/8 to 4/4 (ablation) | LCQ W3/A3, ResNet-20 · CIFAR-10 | top-1 accuracy (%): 92.8 -> 92.6 | Hurt |
| Reduce outer bit-widths from 8/8 to 4/4 (ablation) | LCQ W3/A3, ResNet-18 · ImageNet | top-1 accuracy (%): 70.6 -> 70.4 | Hurt |
Closest papers by meaning
- Quantization Networks (CVPR 2019 · similarity 0.89)
- Incremental Network Quantization: Towards Lossless CNNs with Low-precision Weights (ICLR 2017 · similarity 0.89)
- Nonuniform-to-Uniform Quantization: Towards Accurate Quantization via Generalized Straight-Through Estimation (CVPR 2022 · similarity 0.89)
- Mr.BiQ: Post-Training Non-Uniform Quantization Based on Minimizing the Reconstruction Error (CVPR 2022 · similarity 0.89)
- ParetoQ: Improving Scaling Laws in Extremely Low-bit LLM Quantization (NeurIPS 2025 · similarity 0.87)
- Efficient Multi-bit Quantization Network Training via Weight Bias Correction and Bit-wise Coreset Sampling (NeurIPS 2025 · similarity 0.87)
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.