Research map / Architectures and efficiency / Quantization
Additive Powers-of-Two Quantization: An Efficient Non-uniform Discretization for Neural Networks
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 APoT instead of PACT | ResNet-50 · ImageNet, 4/4 quantization | Top-1 accuracy: 76.5 -> 76.6 | Helped |
| Use APoT instead of QIL | ResNet-18 · ImageNet, 3/3 quantization | Top-1 accuracy: 69.2 -> 69.9 | Helped |
| Use APoT instead of LQ-Net | ResNet-20 · CIFAR10, 3-bit | Accuracy: 91.6 -> 92.2 | Helped |
| Replace APoT levels with PoT levels (ablation) | ResNet-18 · ImageNet, 5/5 quantization | Top-1 accuracy: 70.9 -> 70.3 | Hurt |
| Replace APoT levels with uniform levels (ablation) | ResNet-18 · ImageNet, 5/5 quantization | Top-1 accuracy: 70.9 -> 70.7 | Hurt |
| Replace APoT levels with uniform levels (ablation) | ResNet-18 · ImageNet, 3/3 quantization | Top-1 accuracy: 69.9 -> 69.4 | Hurt |
| Remove RCF (ablation) | APoT ResNet-18 · ImageNet, 3/3 quantization | Top-1 accuracy: 68.5 -> 69.9 | Hurt |
| Add RCF (ablation) | APoT ResNet-18 · ImageNet, 3/3 quantization without WN | Top-1 accuracy | Hurt |
| Remove weight normalization (ablation) | 3-bit ResNet-20 quantization · CIFAR10 | Accuracy: 91.7 -> 0.2 | Hurt |
| Use weight normalization across clipping-threshold learning rates (ablation) | 3-bit ResNet-20 quantization · CIFAR10 | Accuracy: 91.6, 91.7, 91.6, 91.8 vs 0.2, 0.2, 62.8, 84.7 | Helped |
| Use RCF-learned clipping rather than no clipping (ablation) | APoT ResNet-18 · ImageNet, 3/3 quantization | Top-1 accuracy: 68.5 -> 69.9 | Helped |
Closest papers by meaning
- Nonuniform-to-Uniform Quantization: Towards Accurate Quantization via Generalized Straight-Through Estimation (CVPR 2022 · similarity 0.89)
- PowerQuant: Automorphism Search for Non-Uniform Quantization (ICLR 2023 · similarity 0.87)
- Weighted-Entropy-Based Quantization for Deep Neural Networks (CVPR 2017 · similarity 0.86)
- Gradient $\ell_1$ Regularization for Quantization Robustness (ICLR 2020 · similarity 0.86)
- Incremental Network Quantization: Towards Lossless CNNs with Low-precision Weights (ICLR 2017 · similarity 0.86)
- Learnable Companding Quantization for Accurate Low-Bit Neural Networks (CVPR 2021 · 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.