Research map / Architectures and efficiency / Quantization
PowerQuant: Automorphism Search for Non-Uniform 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 |
|---|---|---|---|
| use power function quantization instead of uniform (ablation) | PowerQuant · ResNet 50 W4/A4 | top-1 accuracy (%): 54.68 -> 70.29 | Helped |
| use power function quantization instead of logarithmic (ablation) | PowerQuant · ResNet 50 W4/A4 | top-1 accuracy (%): 57.07 -> 70.29 | Helped |
| use power function quantization instead of logarithmic (ablation) | PowerQuant · DenseNet 121 W4/A4 | top-1 accuracy (%): 5.28 -> 68.04 | Helped |
| use power function quantization instead of uniform (ablation) | PowerQuant · DenseNet 121 W4/A4 | top-1 accuracy (%): 54.83 -> 68.04 | Helped |
| use power function quantization instead of uniform (ablation) | PowerQuant · ResNet 50 W8/A8 | top-1 accuracy (%): 76.15 -> 76.15 | No clear effect |
| use power function quantization instead of uniform (ablation) | PowerQuant · ResNet 50 W6/A6 | top-1 accuracy (%): 75.07 -> 75.95 | Helped |
| fit exponent a globally instead of per layer (ablation) | PowerQuant · ResNet 50 W4/A4 | top-1 accuracy (%): 64,19 -> 70.29 | Helped |
| fit exponent a globally instead of per layer (ablation) | PowerQuant · ResNet 50 W8/A8 | top-1 accuracy (%): 76.14 -> 76.15 | No clear effect |
| choose Nelder-Mead versus Powell or COBYLA solver (ablation) | PowerQuant · MobileNet V2 W6/A6 | top-1 accuracy (%): 64.248 vs 64.104 and 64.364 | No clear effect |
| PowerQuant versus SQuant, data-free | PowerQuant · ResNet 50 W4/A4 | top-1 accuracy (%): 68.60 -> 70.53 | Helped |
| PowerQuant versus SQuant on transformers | PowerQuant · ViT Base W4/A8 | top-1 accuracy (%): 64.62 -> 75.24 | Helped |
| PowerQuant versus PSAQ | PowerQuant · DeiT Tiny W4/A8 | top-1 accuracy (%): 65.57 -> 69.77 | Helped |
Closest papers by meaning
- Nonuniform-to-Uniform Quantization: Towards Accurate Quantization via Generalized Straight-Through Estimation (CVPR 2022 · similarity 0.88)
- Don’t Waste Your Bits! Squeeze Activations and Gradients for Deep Neural Networks via TinyScript (ICML 2020 · similarity 0.88)
- MetaQuant: Learning to Quantize by Learning to Penetrate Non-differentiable Quantization (NeurIPS 2019 · similarity 0.87)
- Additive Powers-of-Two Quantization: An Efficient Non-uniform Discretization for Neural Networks (ICLR 2020 · similarity 0.87)
- Efficient and Generalizable Mixed-Precision Quantization via Topological Entropy (NeurIPS 2025 · similarity 0.87)
- Quantization without Tears (CVPR 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.