Research map / Architectures and efficiency / Quantization
Post-training Quantization with Multiple Points: Mixed Precision without Mixed Precision
Read the paperCheck 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 |
|---|---|---|---|
| add multipoint quantization (ablation) | per-layer quantized ImageNet models · ImageNet, ResNet-101, W4A8 | Top-1 accuracy: 61.04 -> 73.09 | Helped |
| add multipoint quantization (ablation) | per-layer quantized ImageNet models · ImageNet, VGG19-BN, W4A8 | Top-1 accuracy: 60.81 -> 64.06 | Helped |
| add multipoint quantization (ablation) | SSD-VGG16 per-layer quantization · PASCAL VOC, W3A8 | mAP: 42.56% -> 46.94% | Helped |
| add optimal weight clipping (ablation) | multipoint quantization · ImageNet, MobileNet-v2, per-channel W4A4 | Top-1 accuracy: 27.52 -> 55.54 | Helped |
| add optimal weight clipping (ablation) | multipoint quantization · ImageNet, ResNet-101, per-channel W4A4 | Top-1 accuracy: 71.56 -> 72.85 | Helped |
| add optimal weight clipping (ablation) | multipoint quantization · ImageNet, VGG19-BN, per-channel W4A4 | Top-1 accuracy: 71.96 -> 72.78 | Helped |
| add multipoint quantization (ablation) | per-channel quantized ImageNet models · ImageNet, ResNet-101, W4A4 | Top-1 accuracy: 74.22 -> 72.85 | Hurt |
| add multipoint quantization | per-channel quantized ImageNet models · ImageNet, VGG19-BN, W4A4 | Top-1 accuracy: 70.59 -> 72.78 | Helped |
| choose channels by output error instead of randomly | multipoint quantization of ResNet-101 · ImageNet, per-layer W4A8 | accuracy | Helped |
Closest papers by meaning
- Quantization Networks (CVPR 2019 · similarity 0.88)
- Incremental Network Quantization: Towards Lossless CNNs with Low-precision Weights (ICLR 2017 · similarity 0.88)
- Zero-Shot Quantization via Weight-Space Arithmetic (NeurIPS 2026 · similarity 0.87)
- Reg-PTQ: Regression-specialized Post-training Quantization for Fully Quantized Object Detector (CVPR 2024 · similarity 0.86)
- Efficient Multi-bit Quantization Network Training via Weight Bias Correction and Bit-wise Coreset Sampling (NeurIPS 2025 · similarity 0.86)
- MagR: Weight Magnitude Reduction for Enhancing Post-Training Quantization (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.