Research map / Architectures and efficiency / RNNs and normalization
Temporal Pyramid Recurrent Neural Network
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 |
|---|---|---|---|
| add iterative aggregation connections (shortcut path) (ablation) | TP-RNN-S · pMNIST | accuracy (%): 95.1 -> 96.0 | Helped |
| add iterative aggregation connections (ablation) | TP-RNN-S · MNIST | accuracy (%): 97.8 -> 98.7 | Helped |
| add top-down connections between sub-pyramids (ablation) | TP-RNN-S · MNIST | accuracy (%): 98.4 -> 98.7 | Helped |
| add top-down connections (ablation) | TP-RNN-M · pMNIST | accuracy (%): 96.1 -> 96.7 | Helped |
| add output aggregation over all layers (ablation) | TP-RNN-M · MNIST | accuracy (%): 99.0 -> 99.2 | Helped |
| add output aggregation over all layers (ablation) | TP-RNN-M · pMNIST | accuracy (%): 96.2 -> 96.7 | Helped |
| use fine-grained attention aggregation instead of SENet-style aggregation (ablation) | TP-RNN-S · pMNIST | accuracy (%): 95.6 -> 96.0 | Helped |
| use attention aggregation instead of mean pooling (ablation) | TP-RNN-S · MNIST | accuracy (%): 97.8 -> 98.7 | Helped |
| use max pooling instead of SENet aggregation (ablation) | TP-RNN-M · pMNIST | accuracy (%): 96.2 vs 96.0 | Mixed |
| replace LSTM with TP-RNN-S at similar parameters | LSTM-S · pMNIST | accuracy (%): 91.3 -> 96.0 | Helped |
| use TP-RNN-M instead of Dilated LSTM | Dilated LSTM · Signal Frequency Counting | accuracy (%): 76.7 -> 99.8 | Helped |
| stack layers (multi-layer LSTM-M vs LSTM-S) | LSTM · Signal Type Identification | accuracy (%): 92.8 -> 43.7 | Hurt |
Closest papers by meaning
- UnICORNN: A recurrent model for learning very long time dependencies (ICML 2021 · similarity 0.88)
- Shallow RNN: Accurate Time-series Classification on Resource Constrained Devices (NeurIPS 2019 · similarity 0.88)
- Making Convolutional Networks Recurrent for Visual Sequence Learning (CVPR 2018 · similarity 0.87)
- Learning Longer-term Dependencies in RNNs with Auxiliary Losses (ICML 2018 · similarity 0.87)
- Fast-Slow Recurrent Neural Networks (NeurIPS 2017 · similarity 0.87)
- Independently Recurrent Neural Network (IndRNN): Building a Longer and Deeper RNN (CVPR 2018 · 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.