Research map / Architectures and efficiency / Quantization
PD-Quant: Post-Training Quantization Based on Prediction Difference Metric
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 |
|---|---|---|---|
| Use PD-Quant instead of QDrop | ResNet-18 · ImageNet val, W2A2 | top-1 accuracy (%): 51.42 -> 53.14 | Helped |
| Use PD-Quant instead of QDrop | MobileNetV2 · ImageNet val, W2A2 | top-1 accuracy (%): 10.28 -> 13.76 | Helped |
| Use PD-Quant instead of QDrop | ResNet-18 and MobileNetV2 · ImageNet val, W4A4 | top-1 accuracy (%): 69.17 -> 69.23±0.06; 68.07 -> 68.19±0.12 | Helped |
| Optimize quantization parameters using PD loss only, without regularization or DC (ablation) | ResNet-18 · ImageNet val, W2A2 | top-1 accuracy (%): 51.42 -> 1.07 | Hurt |
| Add regularization to PD loss optimization (ablation) | ResNet-18 · ImageNet val, W2A2 | top-1 accuracy (%): 1.07 -> 52.74 | Helped |
| Add regularization to PD loss optimization (ablation) | MobileNetV2 · ImageNet val, W2A2 | top-1 accuracy (%): 7.01 -> 13.49 | Helped |
| Add Distribution Correction to QDrop (ablation) | ResNet-18 · ImageNet val, W2A2 | top-1 accuracy (%): 51.42 -> 52.32 | Helped |
| Add Distribution Correction to QDrop (ablation) | MobileNetV2 · ImageNet val, W2A2 | top-1 accuracy (%): 10.28 -> 10.38 | Helped |
| Add Distribution Correction to PD+Reg (ablation) | MobileNetV2 · ImageNet val, W2A2 | top-1 accuracy (%): 13.49 -> 14.17 | Helped |
| Use KL divergence rather than global MSE or cosine for PD loss (ablation) | ResNet-18 and RegNetX-600MF activation scaling · ImageNet val, W8A2 and W4A2 | top-1 accuracy (%): MSE^g: 27.06, 11.13, 3.70, 1.64; Cosine^g: 27.28, 7.12, 5.93, 3.06; KL^g: 28.41, 12.27, 7.47, 3.17 | Helped |
Closest papers by meaning
- Enhancing Post-training Quantization Calibration through Contrastive Learning (CVPR 2024 · similarity 0.89)
- 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 0.89)
- Predicting Quantization Price for Selecting PTQ Configurations Before Deployment (NeurIPS 2026 · similarity 0.87)
- Reg-PTQ: Regression-specialized Post-training Quantization for Fully Quantized Object Detector (CVPR 2024 · similarity 0.86)
- Evaluating Quantized Large Language Models (ICML 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.