Research map / Architectures and efficiency / Quantization
Mr.BiQ: Post-Training Non-Uniform Quantization Based on Minimizing the Reconstruction Error
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 |
|---|---|---|---|
| jointly optimize multi-level scaling factors and binary codes (ablation) | Mr.BiQ · ImageNet, ResNet-18 | top-1 accuracy (%): 37.20 vs 67.92 | Helped |
| add binary-code optimization to scaling-factor-only optimization (ablation) | Mr.BiQ · ImageNet, ResNet-18 | top-1 accuracy (%): 65.23 vs 70.17 | Helped |
| add scaling-factor optimization to binary-code-only optimization (ablation) | Mr.BiQ · ImageNet, ResNet-18 | top-1 accuracy (%): 66.83 vs 67.92 | Helped |
| use bottom-up joint quantization-parameter optimization instead of BRECQ | ImageNet ResNet-18 W2A4 · ImageNet | top-1 accuracy (%): 64.80±0.08 vs 66.61±0.10 | Helped |
| use bottom-up joint quantization-parameter optimization instead of BRECQ | ImageNet MobileNetV2 W2A4 · ImageNet | top-1 accuracy (%): 53.34±0.15 vs 57.27±0.22 | Helped |
| use bottom-up joint quantization-parameter optimization instead of BRECQ | ImageNet RegNetX-3.2GF W2A32 · ImageNet | top-1 accuracy (%): 73.88±0.14 vs 74.89±0.10 | Helped |
| use Mr.BiQ rather than BRECQ for 2-bit weights | DeiT-S · ImageNet | top-1 accuracy (%): 68.92±0.15 vs 73.15±0.16 | Helped |
| use Mr.BiQ rather than BRECQ for 2-bit weights | DistilBERT · SQuAD v1.1 | F1: 80.39±0.12 vs 83.78±0.16 | Helped |
| use 2-bit Mr.BiQ instead of TernaryBERT | BERT · MRPC | F1: 90.44 vs 89.13±0.36 | Hurt |
| optimize scaling factors only rather than both scaling factors and codes (ablation) | Mr.BiQ W4A32 · ImageNet, ResNet-18 | top-1 accuracy (%): 68.88 vs 70.76 | Hurt |
| optimize binary codes only rather than both codes and scaling factors (ablation) | Mr.BiQ W4A32 · ImageNet, ResNet-18 | top-1 accuracy (%): 70.41 vs 70.76 | Hurt |
Closest papers by meaning
- PTMQ: Post-training Multi-Bit Quantization of Neural Networks (AAAI 2024 · similarity 0.89)
- Exploring Post-training Quantization in LLMs from Comprehensive Study to Low Rank Compensation (AAAI 2024 · similarity 0.88)
- QBB: Quantization with Binary Bases for LLMs (NeurIPS 2024 · similarity 0.87)
- APHQ-ViT: Post-Training Quantization with Average Perturbation Hessian Based Reconstruction for Vision Transformers (CVPR 2025 · similarity 0.87)
- CBQ: Cross-Block Quantization for Large Language Models (ICLR 2025 · similarity 0.87)
- AffineQuant: Affine Transformation Quantization for Large Language Models (ICLR 2024 · 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.