Research map / Architectures and efficiency / RNNs and normalization
Towards Extremely Compact RNNs for Video Recognition With Fully Decomposed Hierarchical Tucker Structure
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 |
|---|---|---|---|
| Integrate the LSTM weight matrices, then apply HT decomposition to the full model (ablation) | HT-LSTM · UCF11, end-to-end direct training | Top-1 accuracy (%): 87.2 -> 87.5 | Helped |
| Integrate the LSTM weight matrices, then apply HT decomposition to the full model (ablation) | CNN + HT-LSTM · UCF11, pre-trained CNN | Top-1 accuracy (%): 98.1 -> 98.4 | Helped |
| Integrate the LSTM weight matrices, then apply HT decomposition to the full model (ablation) | CNN + HT-LSTM · HMDB51, pre-trained CNN | Top-1 accuracy (%): 64.2 -> 64.2 | No clear effect |
| Apply HT decomposition separately to each RNN weight layer instead of integrating layers first (ablation) | HT-LSTM · UCF11 | accuracy: 2.4% accuracy drop | Hurt |
| Use full HT decomposition rather than TT decomposition | LSTM · UCF11, end-to-end direct training | Top-1 accuracy (%): 79.6 -> 87.5 | Helped |
| Use full HT decomposition rather than BT decomposition | LSTM · UCF11, end-to-end direct training | Top-1 accuracy (%): 85.3 -> 87.5 | Helped |
| Use full HT decomposition rather than TR decomposition | LSTM · UCF11, end-to-end direct training | Top-1 accuracy (%): 86.9 -> 87.5 | Helped |
| Use full HT decomposition rather than TT decomposition | LSTM · Youtube Celebrities Face, end-to-end direct training | Top-1 accuracy (%): 75.5 -> 88.2 | Helped |
| Use full HT decomposition rather than TR decomposition | CNN + LSTM · UCF11, pre-trained CNN | Top-1 accuracy (%): 93.8 -> 98.4 | Helped |
| Use full HT decomposition rather than TR decomposition | CNN + LSTM · HMDB51, pre-trained CNN | Top-1 accuracy (%): 63.8 -> 64.2 | Helped |
Closest papers by meaning
- Learning Compact Recurrent Neural Networks With Block-Term Tensor Decomposition (CVPR 2018 · similarity 0.91)
- Compressing Recurrent Neural Networks with Tensor Ring for Action Recognition (AAAI 2019 · similarity 0.90)
- Tensor-Train Recurrent Neural Networks for Video Classification (ICML 2017 · similarity 0.88)
- Towards Efficient Tensor Decomposition-Based DNN Model Compression With Optimization Framework (CVPR 2021 · similarity 0.85)
- Recurrent Neural Networks With Intra-Frame Iterations for Video Deblurring (CVPR 2019 · similarity 0.85)
- Improving Bilinear RNN with Closed-loop Control (NeurIPS 2025 · similarity 0.83)
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.