Research map / Architectures and efficiency / Quantization
Searching for Low-Bit Weights in Quantized 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 SLB instead of DSQ for binary weights and activations | ResNet20 · CIFAR-10 | accuracy: 84.1 -> 85.5 | Helped |
| use SLB instead of LQ-Net with 1-bit weights and 2-bit activations | ResNet20 · CIFAR-10 | accuracy: 88.4 -> 89.5 | Helped |
| use SLB instead of LQ-Net with 1-bit weights and 2-bit activations | VGG-Small · CIFAR-10 | accuracy: 93.4 -> 93.4 | No clear effect |
| use SLB instead of BONN for binary weights and activations | ResNet18 · ILSVRC2012 | Top-1 accuracy: 59.3 -> 61.3 | Helped |
| use SLB instead of LQ-Net with 1-bit weights and 2-bit activations | ResNet18 · ILSVRC2012 | Top-1 accuracy: 62.6 -> 64.8 | Helped |
| use SLB instead of LQ-Net with 2-bit weights and 32-bit activations | ResNet18 · ILSVRC2012 | Top-1 accuracy: 68.0 -> 68.4 | Helped |
| add state batch normalization to SLB (ablation) | ResNet18 with binary weights and activations · ILSVRC2012 | Top-1 accuracy: 61.0 -> 61.3 | Helped |
| use an exponential rather than sine or linear temperature scheduler (ablation) | Binary ResNet20 · CIFAR-10 | accuracy | Helped |
| replace the proposed SLB binarization with DoReFa | VDSR · Set5, ×2 and ×3 super-resolution | PSNR | Hurt |
Closest papers by meaning
- Quantization Networks (CVPR 2019 · similarity 0.90)
- Efficient Quantization for Neural Networks with Binary Weights and Low Bitwidth Activations (AAAI 2019 · similarity 0.89)
- Analysis of Quantized Models (ICLR 2019 · similarity 0.88)
- Linear Symmetric Quantization of Neural Networks for Low-precision Integer Hardware (ICLR 2020 · similarity 0.88)
- Training Quantized Nets: A Deeper Understanding (NeurIPS 2017 · similarity 0.88)
- Weighted-Entropy-Based Quantization for Deep Neural Networks (CVPR 2017 · 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.