Research map / Architectures and efficiency / Quantization
Loss-aware Weight Quantization of Deep 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 loss-aware ternarization instead of TWN | LSTM · War and Peace | testing cross-entropy: 1.290 -> 1.245 | Helped |
| Use loss-aware ternarization instead of TTQ | LSTM · Linux Kernel | testing cross-entropy: 1.302 -> 1.256 | Helped |
| Use loss-aware ternarization instead of full precision | LSTM · Penn Treebank | testing cross-entropy: 1.083 -> 1.015 | Helped |
| Use loss-aware 3-bit logarithmic quantization instead of DoReFa-Net3 | LSTM · Penn Treebank | testing cross-entropy: 1.017 -> 1.009 | Helped |
| Use two scaling parameters instead of one (ablation) | LAT ternarization · CIFAR-100 | testing error (%): 39.19 -> 38.84 | Helped |
| Use two scaling parameters instead of one (ablation) | LAT ternarization · War and Peace | testing cross-entropy: 1.253 -> 1.245 | Helped |
| Use two scaling parameters instead of one (ablation) | LAT ternarization · CIFAR-10 | testing error (%): 10.38 -> 10.48 | Hurt |
| Use two scaling parameters instead of one (ablation) | LAT ternarization · SVHN | testing error (%): 2.30 -> 2.35 | Mixed |
| Use logarithmic rather than linear quantization (ablation) | LAQ3 · Penn Treebank | testing cross-entropy: 1.017 -> 1.009 | Helped |
| Use logarithmic rather than linear quantization (ablation) | LAQ3 · CIFAR-100 | testing error (%): 38.70 -> 38.50 | Helped |
| Use four bits rather than ternarization (ablation) | LAQ logarithmic quantization · War and Peace | testing cross-entropy: 1.239 -> 1.272 | Hurt |
| Use four bits rather than ternarization (ablation) | LAQ logarithmic quantization · Linux Kernel | testing cross-entropy: 1.256 -> 1.319 | Hurt |
Closest papers by meaning
- Adaptive Loss-Aware Quantization for Multi-Bit Networks (CVPR 2020 · similarity 0.91)
- Explicit Loss-Error-Aware Quantization for Low-Bit Deep Neural Networks (CVPR 2018 · similarity 0.91)
- Loss-aware Binarization of Deep Networks (ICLR 2017 · similarity 0.90)
- One-Step Forward and Backtrack: Overcoming Zig-Zagging in Loss-Aware Quantization Training (AAAI 2024 · similarity 0.89)
- Analysis of Quantized Models (ICLR 2019 · similarity 0.89)
- Weighted-Entropy-Based Quantization for Deep Neural Networks (CVPR 2017 · 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.