Research map / Architectures and efficiency / Quantization
Finding Non-Uniform Quantization Schemes using Multi-Task Gaussian Processes
Marcelo Gennari do Nascimento, Theo W. Costain, Victor Adrian Prisacariu
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (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 decreasing bit-width (more bits early) instead of inverted order (ablation) | VGG16 · CIFAR10 | Top-1 (%): 87.7 -> 93.7 | Improved |
| use decreasing bit-width instead of inverted order (ablation) | VGG19 · CIFAR10 | Top-1 (%): 89.6 -> 93.7 | Improved |
| use decreasing bit-width instead of inverted order (ablation) | ResNet18 · CIFAR10 | Top-1 (%): 92.9 -> 95.4 | Improved |
| use decreasing bit-width instead of inverted order (ablation) | GoogLeNet · CIFAR10 | Top-1 (%): 94.2 -> 95.3 | Improved |
| use decreasing bit-width instead of inverted order (ablation) | VGG16 · CIFAR10 | Memory (MB): 9.50 -> 4.84 | Improved |
| use searched decreasing bits instead of uniform 3-bit (ablation) | VGG16 · CIFAR10 | Memory (MB): 6.44 -> 4.84 | Improved |
| use searched decreasing bits instead of uniform 3-bit (ablation) | ResNet18 · CIFAR10 | Top-1 (%): 95.0 -> 95.4 | Improved |
| use searched bits instead of uniform 3-bit | GoogLeNet · CIFAR10 | Top-1 (%): 95.1 -> 95.3 | Mixed |
| use searched decreasing bits instead of uniform 2-bit (ablation) | VGG19 · CIFAR10 | Top-1 (%): 92.2 -> 93.7 | Improved |
| use Chebyshev 4th-degree basis instead of Bezier linear (ablation) | ResNet18 · CIFAR10 | Accuracy loss (%) | No clear effect |
| search with ours vs ReLeQ on VGG16 | VGG16 · CIFAR10 | Model size (MB): 13.32 -> 4.62 | Improved |
| search with ours vs DSConv uniform 4-bit | ResNet50 · ImageNet | Model size (MB): 14.54 -> 11.89 | Improved |
Closest papers by meaning
- Once Quantization-Aware Training: High Performance Extremely Low-Bit Architecture Search (ICCV 2021 · similarity 0.88)
- Nonuniform-to-Uniform Quantization: Towards Accurate Quantization via Generalized Straight-Through Estimation (CVPR 2022 · similarity 0.88)
- Finding the Task-Optimal Low-Bit Sub-Distribution in Deep Neural Networks (ICML 2022 · similarity 0.88)
- Quantization without Tears (CVPR 2025 · similarity 0.87)
- Deep Neural Network Quantization via Layer-Wise Optimization Using Limited Training Data (AAAI 2019 · similarity 0.87)
- RDO-Q: Extremely Fine-Grained Channel-Wise Quantization via Rate-Distortion Optimization (ECCV 2022 · 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.
