Research map / Architectures and efficiency / Quantization
OMPQ: Orthogonal Mixed Precision Quantization
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 |
|---|---|---|---|
| Allocate mixed weight bit-widths using ORM-based linear programming | ResNet-18 QAT · ImageNet | Top-1 accuracy: 70.22 -> 72.08 | Helped |
| Allocate mixed weight bit-widths using ORM-based linear programming | ResNet-50 QAT · ImageNet | Top-1 accuracy: 75.39 -> 76.28 | Helped |
| Replace BRECQ evolutionary search with ORM-based allocation | ResNet-18 PTQ · ImageNet | Top-1 accuracy: 68.82 -> 69.41 | Helped |
| Replace BRECQ evolutionary search with ORM-based allocation | MobileNetV2 PTQ · ImageNet | Top-1 accuracy: 70.28 -> 71.39 | Helped |
| Use −logx instead of e−x as the decreasing function (ablation) | OMPQ · ImageNet, ResNet-18 QAT | Top-1 accuracy: 72.30 -> 72.26 | No clear effect |
| Use −x2 instead of e−x as the decreasing function (ablation) | OMPQ · ImageNet, ResNet-18 QAT | Top-1 accuracy: 72.30 -> 72.36 | No clear effect |
| Use −x3 instead of e−x as the decreasing function (ablation) | OMPQ · ImageNet, ResNet-18 QAT | Top-1 accuracy: 72.30 -> 71.71 | Hurt |
| Use a faster-changing decreasing function rather than e−x (ablation) | OMPQ · ImageNet, MobileNetV2 PTQ | Top-1 accuracy: 63.51 -> 63.20; 63.0 | Hurt |
| Use layer-wise rather than block-wise deconstruction (ablation) | OMPQ · ImageNet, ResNet-18 QAT | Top-1 accuracy: 72.52 -> 72.51 | No clear effect |
| Use finer layer-wise rather than net-wise deconstruction (ablation) | OMPQ · ImageNet, ResNet-18 QAT | Top-1 accuracy: 72.31 -> 72.51 | No clear effect |
| Use finer layer-wise rather than block-wise deconstruction (ablation) | OMPQ · ImageNet, MobileNetV2 PTQ | Top-1 accuracy: 69.10 -> 69.37 | No clear effect |
| Use finer layer-wise rather than stage-wise deconstruction (ablation) | OMPQ · ImageNet, MobileNetV2 PTQ | Top-1 accuracy: 68.86 -> 69.37 | Helped |
Closest papers by meaning
- InfoQ: Mixed-Precision Quantization via Global Information Flow (AAAI 2026 · similarity 0.90)
- HAWQ-V3: Dyadic Neural Network Quantization (ICML 2021 · similarity 0.89)
- One-Shot Model for Mixed-Precision Quantization (CVPR 2023 · similarity 0.88)
- Retraining-Free Model Quantization via One-Shot Weight-Coupling Learning (CVPR 2024 · similarity 0.86)
- MetaMix: Meta-State Precision Searcher for Mixed-Precision Activation Quantization (AAAI 2024 · similarity 0.86)
- PTMQ: Post-training Multi-Bit Quantization of Neural Networks (AAAI 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.