Research map / Architectures and efficiency / Quantization
AdaBits: Neural Network Quantization With Adaptive Bit-Widths
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 |
|---|---|---|---|
| jointly train multiple bit-widths with shared weights | ResNet50 · ImageNet | top-1 accuracy: 73.2 vs 29.5 | Helped |
| use independent clipping levels for each bit-width (S-CL) (ablation) | Vanilla AdaBits MobileNet V1 · ImageNet, 4-bit | top-1 accuracy: 70.8 -> 71.1 | Helped |
| share clipping levels across bit-widths instead of using S-CL (ablation) | AdaBits MobileNet V1 · ImageNet, 4-bit | top-1 accuracy: 71.1 -> 70.8 | Hurt |
| train progressively from low to high bit-width (ablation) | ResNet50 · ImageNet, 2-bit | top-1 accuracy: 73.3 -> 29.5 | Hurt |
| train progressively from high to low bit-width (ablation) | ResNet50 · ImageNet, 4-bit and 3-bit | top-1 accuracy: 76.3 -> 73.9; 75.9 -> 73.6 | Hurt |
| directly adapt weights trained at 4-bit to lower precision (ablation) | ResNet50 · ImageNet, 2-bit | top-1 accuracy: 73.3 -> 20.3 | Hurt |
| directly adapt weights trained at 2-bit to higher precision (ablation) | ResNet50 · ImageNet, 4-bit | top-1 accuracy: 76.3 -> 73.4 | Hurt |
| use modified floor-based quantization that permits dropping lower weight bits (ablation) | AdaBits MobileNet V1 and MobileNet V2 · ImageNet | model size: 4.35 MB; 3.83 MB | Helped |
Closest papers by meaning
- Adaptive Quantization for Deep Neural Network (AAAI 2018 · similarity 0.89)
- Distribution-Aware Adaptive Multi-Bit Quantization (CVPR 2021 · similarity 0.88)
- Fixed-Point Back-Propagation Training (CVPR 2020 · similarity 0.88)
- Adaptive Loss-Aware Quantization for Multi-Bit Networks (CVPR 2020 · similarity 0.88)
- PTMQ: Post-training Multi-Bit Quantization of Neural Networks (AAAI 2024 · similarity 0.87)
- Efficient Multi-bit Quantization Network Training via Weight Bias Correction and Bit-wise Coreset Sampling (NeurIPS 2025 · similarity 0.87)
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.