Research map / Architectures and efficiency / Quantization
Deep Neural Network Quantization via Layer-Wise Optimization Using Limited Training Data
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 |
|---|---|---|---|
| add weights update of remaining layers during layer-wise quantization (ablation) | L-DNQ · CIFAR-10 ResNet20, 3-bit | accuracy improvement over FP (%): -13.66 -> -4.3 | Helped |
| add weights update of remaining layers (ablation) | L-DNQ · ImageNet AlexNet top-1, 3-bit | accuracy improvement over FP (%): -56.80 -> -17.78 | Helped |
| increase quantization bits from 3 to 9 (ablation) | L-DNQ · ImageNet ResNet18 top-1, 1% data | accuracy improvement over FP (%): -16.43 -> -2.73 | Helped |
| use L-DNQ instead of training-based quantization (TTQ) with 1% data | ResNet20 · CIFAR-10, 3-bit | accuracy improvement over FP (%): -77.25 -> -4.30 | Helped |
| use L-DNQ instead of direct quantization DQ | ResNet18 · ImageNet top-1, 3-bit | accuracy improvement over FP (%): -61.22 -> -16.43 | Helped |
| lower final layer output error between quantized and original network | quantized ResNet18 · ImageNet | test error | Helped |
| use L-DNQ 9-bit instead of DQ with 8 to 32 bits | ResNet18 · ImageNet top-1 | accuracy improvement over FP (%): -2.82 -> -2.73 | Helped |
Closest papers by meaning
- Adaptive Quantization for Deep Neural Network (AAAI 2018 · similarity 0.92)
- Explicit Loss-Error-Aware Quantization for Low-Bit Deep Neural Networks (CVPR 2018 · similarity 0.90)
- Incremental Network Quantization: Towards Lossless CNNs with Low-precision Weights (ICLR 2017 · similarity 0.89)
- Fixed-Point Back-Propagation Training (CVPR 2020 · similarity 0.89)
- Quantization without Tears (CVPR 2025 · similarity 0.89)
- Layer-wise Quantization for Quantized Optimistic Dual Averaging (ICML 2025 · 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.