Research map / Architectures and efficiency / Quantization
Normalization Helps Training of Quantized LSTM
Lu Hou, Jinhua Zhu, James Kwok, Fei Gao, Tao Qin, Tie-Yan Liu
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 |
|---|---|---|---|
| add layer norm (ablation) | BinaryConnect LSTM · Penn Treebank char-level | test BPC: 2.51 -> 1.49 | Improved |
| add layer norm (ablation) | TerConnect LSTM · Penn Treebank char-level | test BPC: 5.84 -> 1.43 | Improved |
| add layer norm (ablation) | BinaryConnect LSTM · War and Peace char-level | test BPC: 4.24 -> 1.69 | Improved |
| add weight norm (ablation) | BinaryConnect LSTM · Penn Treebank word-level d=300 | test perplexity: 8247.4 -> 87.6 | Improved |
| add weight norm (ablation) | TerConnect LSTM · Penn Treebank word-level d=300 | test perplexity: 113.8 -> 86.5 | Improved |
| add layer norm (ablation) | BinaryConnect LSTM · Penn Treebank word-level d=650 | test perplexity: 1244.2 -> 82.3 | Improved |
| add layer norm (ablation) | BinaryConnect LSTM · pMNIST | test accuracy (%): N/A -> 91.2 | Improved |
| add layer norm (ablation) | LAT 2-bit LSTM · Penn Treebank char-level | test BPC: 1.48 -> 1.40 | Improved |
| use shared instead of separate batch norm statistics (ablation) | BinaryConnect LSTM · Penn Treebank char-level | test BPC: 1.50 -> 1.51 | No clear effect |
| use shared instead of separate batch norm statistics (ablation) | BinaryConnect LSTM · pMNIST | test accuracy (%): 91.2 -> 36.5 | Hurt |
| add normalization to full-precision LSTM (layer norm) | Full-precision LSTM · Penn Treebank char-level | test BPC: 1.45 -> 1.43 | Improved |
| apply normalization directly on quantization instead of SBN stochastic binarization with batch norm | BinaryConnect LSTM · Penn Treebank char-level | test BPC: 1.60 -> 1.49 | Improved |
Closest papers by meaning
- Gradient $\ell_1$ Regularization for Quantization Robustness (ICLR 2020 · similarity 0.88)
- Scheduling Weight Transitions for Quantization-Aware Training (ICCV 2025 · similarity 0.88)
- xLSTM: Extended Long Short-Term Memory (NeurIPS 2024 · similarity 0.87)
- MagR: Weight Magnitude Reduction for Enhancing Post-Training Quantization (NeurIPS 2024 · similarity 0.87)
- QBB: Quantization with Binary Bases for LLMs (NeurIPS 2024 · similarity 0.87)
- LQER: Low-Rank Quantization Error Reconstruction for LLMs (ICML 2024 · similarity 0.86)
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.
