Research map / Architectures and efficiency / Quantization
Distance-Aware Quantization
Dohyung Kim, Junghyup Lee, Bumsub Ham
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 |
|---|---|---|---|
| replace fixed temperature beta=4 with adaptive temperature beta* (ablation) | DASR (ResNet-20) · CIFAR-10 test (W/A 1/1) | top-1 accuracy (%): 13.5 -> 85.8 | Improved |
| replace fixed temperature beta=12 with adaptive temperature (ablation) | DASR (ResNet-20) · CIFAR-10 test (W/A 1/1) | top-1 accuracy (%): 69.9 -> 85.8 | Improved |
| use DASR at test time with beta=4 instead of rounding (ablation) | DASR (ResNet-20) · CIFAR-10 test (W/A 1/1) | top-1 accuracy (%): 13.5 -> 89.7 | Improved |
| raise fixed temperature from 12 to 24 (ablation) | DASR (ResNet-20) · CIFAR-10 test (W/A 1/1) | top-1 accuracy (%): 69.9 -> 59.6 | Hurt |
| use kernel soft argmax instead of plain soft argmax at beta=10 vs 12 (ablation) | DASR (ResNet-20) · CIFAR-10 test (W/A 1/1) | top-1 accuracy (%): 11.7 -> 69.9 | Improved |
| replace temperature annealing with temperature controller (ablation) | DAQ (ResNet-20) · CIFAR-10 test (W/A 1/1) | top-1 accuracy (%): 72.2 -> 85.8 | Improved |
| replace combined STE with temperature controller (ablation) | DAQ (ResNet-20) · CIFAR-10 test (W/A 1/1) | top-1 accuracy (%): 81.7 -> 85.8 | Improved |
| DAQ vs best prior at 1/1 ResNet-18 (QNet) | DAQ · ImageNet val ResNet-18 (W/A 1/1) | top-1 accuracy (%): 53.6 -> 56.2 | Improved |
| DAQ vs IRNet at 1/32 ResNet-18 | DAQ · ImageNet val ResNet-18 (W/A 1/32) | top-1 accuracy (%): 66.5 -> 67.2 | Improved |
| DAQ vs LSQ+ at 4/4 ResNet-18 | DAQ · ImageNet val ResNet-18 (W/A 4/4) | top-1 accuracy (%): 70.8 -> 70.5 | No clear effect |
| DAQ vs PROFIT on MobileNet-V2 4/4 | DAQ · ImageNet val MobileNet-V2 (W/A 4/4) | top-1 accuracy (%): 71.6 -> 70.0 | Hurt |
| DAQ vs IRNet at 1/1 ResNet-20 | DAQ · CIFAR-10 test ResNet-20 (W/A 1/1) | top-1 accuracy (%): 85.4 -> 85.8 | Improved |
Closest papers by meaning
- Network Quantization With Element-Wise Gradient Scaling (CVPR 2021 · similarity 0.89)
- LQ-Nets: Learned Quantization for Highly Accurate and Compact Deep Neural Networks (ECCV 2018 · similarity 0.88)
- Up or Down? Adaptive Rounding for Post-Training Quantization (ICML 2020 · similarity 0.88)
- A Statistical Framework for Low-bitwidth Training of Deep Neural Networks (NeurIPS 2020 · similarity 0.87)
- Learning to Quantize Deep Networks by Optimizing Quantization Intervals With Task Loss (CVPR 2019 · similarity 0.87)
- Understanding the Unfairness in Network Quantization (ICML 2025 · 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.
