Research map / Architectures and efficiency / Quantization
Extremely Low Bit Neural Network: Squeeze the Last Bit Out With ADMM
Cong Leng, Zesheng Dou, Hao Li, Shenghuo Zhu, Rong Jin
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 |
|---|---|---|---|
| Use ADMM ternary quantization instead of TWN | AlexNet · ImageNet validation | Top-1 accuracy: 0.575 -> 0.582 | Improved |
| Use ADMM ternary quantization instead of TWN | VGG-16 · ImageNet validation | Top-1 accuracy: 0.691 -> 0.700 | Improved |
| Use ADMM ternary quantization instead of TWN | ResNet-18 · ImageNet validation | Top-1 accuracy: 0.618 -> 0.670 | Improved |
| Use ADMM ternary quantization instead of TWN | ResNet-50 · ImageNet validation | Top-1 accuracy: 0.656 -> 0.725 | Improved |
| Use ADMM ternary quantization instead of TWN | GoogleNet · ImageNet validation | Top-1 accuracy: 0.612 -> 0.631 | Improved |
| Increase weight precision from binary to ternary (ablation) | ImageNet CNNs · ImageNet validation | Top-1 accuracy | Improved |
| Quantize GoogleNet 1×1 kernels with INT8 instead of extremely low bits (ablation) | GoogleNet · ImageNet validation | Top-1 accuracy: 0.603 -> 0.654; 0.631 -> 0.667 | Improved |
| Quantize Darknet 1×1 kernels with INT8 instead of extremely low bits (ablation) | Darknet+SSD · Pascal VOC 2007 test | mAP: 0.609 (0.621); 0.624 (0.639) | Improved |
| Replace full precision weights with {-4, +4} quantized weights (ablation) | VGG-16+SSD · Pascal VOC 2007 test | mAP: 0.778 -> 0.776 | Hurt |
Closest papers by meaning
- Sub-Bit Neural Networks: Learning To Compress and Accelerate Binary Neural Networks (ICCV 2021 · similarity 0.88)
- Quantization Networks (CVPR 2019 · similarity 0.87)
- Deep Neural Network Quantization via Layer-Wise Optimization Using Limited Training Data (AAAI 2019 · similarity 0.87)
- Two-Step Quantization for Low-Bit Neural Networks (CVPR 2018 · similarity 0.87)
- Bit-Shrinking: Limiting Instantaneous Sharpness for Improving Post-Training Quantization (CVPR 2023 · similarity 0.86)
- QBB: Quantization with Binary Bases for LLMs (NeurIPS 2024 · similarity 0.85)
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.
