Research map / Architectures and efficiency / RNNs and normalization
Gating Revisited: Deep Multi-layer RNNs That Can Be Trained
Mehmet Özgür Türkoglu, Stefano D’Aronco, Jan Dirk Wegner, Konrad Schindler B
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 an LSTM-with-forget-gate first layer under 7 STAR layers (ablation) | STAR · pMNIST | accuracy (%): 94.1 -> 95.4 | Improved |
| add an LSTM-with-forget-gate first layer under 7 STAR layers (ablation) | STAR · pixel-by-pixel MNIST | accuracy (%): 99.2 -> 99.4 | Improved |
| increase STAR depth from 8 to 12 layers (ablation) | STAR · pMNIST | accuracy (%): 94.1 -> 94.7 | Improved |
| increase STAR depth from 8 to 12 layers (ablation) | STAR · pixel-by-pixel MNIST | accuracy (%): 99.2 -> 99.2 | No clear effect |
| increase STAR depth from 4 to 6 layers (ablation) | STAR · BreizhCrops | accuracy (%): 68.2 -> 69.6 | Improved |
| increase STAR depth from 4 to 6 layers (ablation) | STAR · TUM time series classification | accuracy (%): 87.7 -> 87.6 | No clear effect |
| increase convSTAR depth from 8 to 12 layers (ablation) | convSTAR · Jester gesture recognition | accuracy (%): 92.3 -> 92.5 | Improved |
| add an LSTM first layer under 7 convSTAR layers (ablation) | convSTAR · Jester gesture recognition | accuracy (%): 92.3 -> 92.7 | Improved |
| add batch normalization between layers (ablation) | IndRNN · pixel-by-pixel MNIST | training stability | Improved |
| replace LSTM with STAR cell | convLSTM (4 layers) · TUM pixel-wise classification | accuracy (%): 90.6 -> 91.9 | Improved |
| replace IndRNN with STAR cell | IndRNN (6 layers) · PennTreebank character-level | bits per character: 1.42 -> 1.30 | Improved |
| replace LSTM with STAR cell by removing output gate and coupling input and forget gates (ablation) | LSTM · 12-layer MNIST gradient flow | gradient magnitude | Improved |
Closest papers by meaning
- Independently Recurrent Neural Network (IndRNN): Building a Longer and Deeper RNN (CVPR 2018 · similarity 0.90)
- Dynamical Isometry and a Mean Field Theory of RNNs: Gating Enables Signal Propagation in Recurrent Neural Networks (ICML 2018 · similarity 0.90)
- Dilated Recurrent Neural Networks (NeurIPS 2017 · similarity 0.89)
- On the Expressivity of Recurrent Neural Cascades (AAAI 2024 · similarity 0.88)
- Recurrent neural networks: vanishing and exploding gradients are not the end of the story (NeurIPS 2024 · similarity 0.88)
- Exploiting Symmetric Temporally Sparse BPTT for Efficient RNN Training (AAAI 2024 · 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.
