Research map / Architectures and efficiency / Quantization
Quantization Networks
Check 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 the differentiable quantization network rather than the strongest listed baseline | AlexNet with 3(±4)-bit weights and 32-bit activations on ImageNet · ImageNet (AlexNet) | Top-1 accuracy: 60.0 -> 61.9 | Helped |
| Use the differentiable quantization network rather than the strongest listed baseline | ResNet-18 with 3(±4)-bit weights and 32-bit activations · ImageNet (ResNet-18) | Top-1 accuracy: 68.0 -> 70.4 | Helped |
| Use the differentiable quantization network rather than LQ-Net | ResNet-50 with 2-bit weights and 32-bit activations · ImageNet (ResNet-50) | Top-1 accuracy: 75.1 -> 75.2 | Helped |
| Use the differentiable quantization network rather than ADMM | SSD with 3(±4)-bit weights and 32-bit activations · Pascal VOC 2007 test | mAP: 77.6 -> 77.7 | Helped |
| Use non-uniform rather than linear quantization intervals (ablation) | AlexNet, 2/32 weights/activations · ImageNet | Top-1 accuracy: $60.6$ -> $60.9$ | Helped |
| Use non-uniform rather than linear quantization intervals (ablation) | AlexNet, 3(±4)/32 weights/activations · ImageNet | Top-1 accuracy: $60.7$ -> $61.9$ | Helped |
| Use layer-wise rather than shared quantization function parameters (ablation) | AlexNet, 2/32 weights/activations · ImageNet | Top-1 accuracy: $59.9$ -> $60.9$ | Helped |
| Fine-tune from a pre-trained model rather than train from scratch (ablation) | ResNet-18, 3(±4)/32 weights/activations · ImageNet | Top-1 accuracy: $55.3$ -> $70.4$ | Helped |
| Increase temperature gradually during training (ablation) | Quantization networks · ImageNet quantization; soft versus hard quantization | accuracy gap | Helped |
| Adaptively learn quantization biases rather than freeze them after initialization (ablation) | Quantization networks · ImageNet quantization | accuracy | No clear effect |
Closest papers by meaning
- Efficient Quantization for Neural Networks with Binary Weights and Low Bitwidth Activations (AAAI 2019 · similarity 0.91)
- Searching for Low-Bit Weights in Quantized Neural Networks (NeurIPS 2020 · similarity 0.90)
- Weighted-Entropy-Based Quantization for Deep Neural Networks (CVPR 2017 · similarity 0.90)
- Don’t Waste Your Bits! Squeeze Activations and Gradients for Deep Neural Networks via TinyScript (ICML 2020 · similarity 0.90)
- Linear Symmetric Quantization of Neural Networks for Low-precision Integer Hardware (ICLR 2020 · similarity 0.90)
- Accurate Neural Training with 4-bit Matrix Multiplications at Standard Formats (ICLR 2023 · similarity 0.89)
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.