Research map / Architectures and efficiency / RNNs and normalization
Learning Compact Recurrent Neural Networks With Block-Term Tensor Decomposition
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 |
|---|---|---|---|
| Replace the dense input-to-hidden transformation with a Block-Term tensor decomposition | LSTM · UCF11 | accuracy: 0.697 -> 0.853 | Helped |
| Use Block-Term rather than Tensor-Train decomposition for the input-to-hidden transformation | LSTM · UCF11 | accuracy: 0.796 -> 0.853 | Helped |
| Replace the dense input-to-hidden transformation with BT decomposition | LSTM · UCF11 | parameters: 58.9 millons -> 3387 | Helped |
| Replace the dense input-to-hidden transformation with BT decomposition | LSTM · UCF11 validation | epochs to reach 60% accuracy: 230 epochs -> epoch-16 | Helped |
| Increase core-order d from 2 to 4 (ablation) | BT-LSTM · Synthetic weight reconstruction | reconstruction quality: Parameters reduce about 1.3 times | Hurt |
| Increase Tucker-rank R from 1 to 4 (ablation) | BT-LSTM · Synthetic weight reconstruction | robustness: from 1 to 4 | Helped |
| Use a proper CP-rank setting with multiple Tucker models (ablation) | BT-LSTM · Synthetic weight reconstruction | robustness to noise | Helped |
| Use BT-LSTM instead of LSTM in the DRAW encoder | DRAW · MNIST image generation | generated-image quality | No clear effect |
| Use BT-LSTM instead of LSTM in the image-captioning model | Neural Image Caption · MSCOCO | caption quality | Helped |
Closest papers by meaning
- Compressing Recurrent Neural Networks with Tensor Ring for Action Recognition (AAAI 2019 · similarity 0.92)
- Tensor-Train Recurrent Neural Networks for Video Classification (ICML 2017 · similarity 0.91)
- Towards Extremely Compact RNNs for Video Recognition With Fully Decomposed Hierarchical Tucker Structure (CVPR 2021 · similarity 0.91)
- Making Convolutional Networks Recurrent for Visual Sequence Learning (CVPR 2018 · similarity 0.88)
- A Tensor Decomposition Perspective on Second-order RNNs (ICML 2024 · similarity 0.86)
- Improving Bilinear RNN with Closed-loop Control (NeurIPS 2025 · 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.