Research map / Architectures and efficiency / RNNs and normalization
Architectural Complexity Measures of Recurrent Neural Networks
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 |
|---|---|---|---|
| increase recurrent depth (bu -> td, d_r 1 -> 2) | tanh RNN · text8 | BPC: 1.80 -> 1.77 | Helped |
| increase recurrent depth d_r 1 -> 3 at d_f=2 | tanh RNN · text8 | BPC: 1.88 -> 1.83 | Helped |
| increase feedforward depth d_f at d_r=3 | tanh RNN · text8 | BPC: 1.83 -> 1.88 | Hurt |
| increase feedforward depth | tanh RNN · sequential MNIST | accuracy | No clear effect |
| increase recurrent depth | tanh RNN · sequential MNIST | accuracy | Hurt |
| increase recurrent skip coefficient s 1 -> 21 | tanh RNN · sequential MNIST | accuracy: 34.9 -> 87.8 | Helped |
| increase recurrent skip coefficient s 1 -> 7 | tanh RNN · pMNIST | accuracy: 49.8 -> 88.9 | Helped |
| skip-connected tanh RNN (stanh) vs uRNN | tanh RNN · pMNIST | accuracy: 91.4 -> 94.0 | Helped |
| add skip connections that do not raise s | stacked tanh RNN · MNIST k=17 | accuracy: 39.5 -> 39.4 | No clear effect |
| arrange skip connections to raise s to k | stacked tanh RNN · pMNIST k=9 | accuracy: 71.1 -> 86.9 | Helped |
| increase recurrent skip coefficient s 1 -> 3 | LSTM · sequential MNIST | accuracy: 56.2 -> 87.2 | Helped |
| increase recurrent skip coefficient s 1 -> 3 | LSTM · pMNIST | accuracy: 28.5 -> 25.0 | Hurt |
Closest papers by meaning
- Capacity and Trainability in Recurrent Neural Networks (ICLR 2017 · similarity 0.89)
- Generalized Tensor Models for Recurrent Neural Networks (ICLR 2019 · similarity 0.88)
- Recurrent neural networks: vanishing and exploding gradients are not the end of the story (NeurIPS 2024 · similarity 0.88)
- The Uncanny Similarity of Recurrence and Depth (ICLR 2022 · similarity 0.88)
- Dilated Recurrent Neural Networks (NeurIPS 2017 · similarity 0.88)
- When Do Transformers Outperform Feedforward and Recurrent Networks? A Statistical Perspective (NeurIPS 2025 · 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.