Research map / Architectures and efficiency / Quantization
SYQ: Learning Symmetric Quantization for Efficient Deep 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 symmetric subgroup scaling instead of the compared low-precision methods | AlexNet with 1-bit weights and 2-bit activations · ImageNet validation | Top-1 accuracy: 52.7 -> 55.4 | Helped |
| Use symmetric subgroup scaling instead of BWN | ResNet-18 with 1-bit weights · ImageNet validation | Top-1 accuracy: 60.8 -> 62.9 | Helped |
| Use symmetric subgroup scaling instead of FGQ | ResNet-50 with 2-bit weights and 4-bit activations · ImageNet validation | Top-1 accuracy: 68.4 -> 70.9 | Helped |
| Use symmetric subgroup scaling instead of FGQ | ResNet-50 with 2-bit weights and 8-bit activations · ImageNet validation | Top-1 accuracy: 70.8 -> 72.3 | Helped |
| Use row-wise instead of pixel-wise scaling (ablation) | SYQ on AlexNet · ImageNet validation, 1-bit weights and 2-bit activations | Top-1 accuracy difference: -0.7 | Hurt |
| Use layer-wise instead of pixel-wise scaling (ablation) | SYQ on AlexNet · ImageNet validation, 1-bit weights and 2-bit activations | Top-1 accuracy difference: -1.4 | Hurt |
| Use row-wise instead of pixel-wise scaling (ablation) | SYQ on AlexNet · ImageNet validation, 1-bit weights and 8-bit activations | Top-1 accuracy difference: -0.1 | Hurt |
| Use layer-wise instead of pixel-wise scaling (ablation) | SYQ on AlexNet · ImageNet validation, 1-bit weights and 8-bit activations | Top-1 accuracy difference: -0.4 | Hurt |
| Use row-wise instead of pixel-wise scaling (ablation) | SYQ on AlexNet · ImageNet validation, 2-bit weights and 2-bit activations | Top-1 accuracy difference: +0.1 | Helped |
| Use layer-wise instead of pixel-wise scaling (ablation) | SYQ on AlexNet · ImageNet validation, 2-bit weights and 2-bit activations | Top-1 accuracy difference: -1.3 | Hurt |
| Use row-wise instead of pixel-wise scaling (ablation) | SYQ on AlexNet · ImageNet validation, 2-bit weights and 8-bit activations | Top-1 accuracy difference: -0.1 | Hurt |
| Use layer-wise instead of pixel-wise scaling (ablation) | SYQ on AlexNet · ImageNet validation, 2-bit weights and 8-bit activations | Top-1 accuracy difference: -1.9 | Hurt |
Closest papers by meaning
- Linear Symmetric Quantization of Neural Networks for Low-precision Integer Hardware (ICLR 2020 · similarity 0.90)
- Analysis of Quantized Models (ICLR 2019 · similarity 0.90)
- Explicit Loss-Error-Aware Quantization for Low-Bit Deep Neural Networks (CVPR 2018 · similarity 0.89)
- Incremental Network Quantization: Towards Lossless CNNs with Low-precision Weights (ICLR 2017 · similarity 0.89)
- Accurate Neural Training with 4-bit Matrix Multiplications at Standard Formats (ICLR 2023 · similarity 0.89)
- QBB: Quantization with Binary Bases for LLMs (NeurIPS 2024 · similarity 0.88)
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.