Research map / Architectures and efficiency / RNNs and normalization
Coupled Oscillatory Recurrent Neural Network (coRNN): An accurate and (gradient) stable architecture for learning long time dependencies
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 |
|---|---|---|---|
| Use coRNN instead of DTRIV∞ | Single-layer RNN · psMNIST | test accuracy: 96.8% -> 97.3% | Helped |
| Use coRNN instead of Lipschitz RNN | RNN for long-term dependencies · Noise padded CIFAR-10 | test accuracy: 55.2% -> 59.0% | Helped |
| Use coRNN instead of incremental RNN | RNN for activity classification · HAR-2 | test accuracy: 96.3% -> 97.2% | Helped |
| Use coRNN instead of LSTM | RNN for sentiment classification · IMDB | test accuracy: 86.8% -> 87.4% | Helped |
| Use 20-unit tiny coRNN instead of 64-unit coRNN (ablation) | coRNN · HAR-2 | test accuracy: 97.2% -> 96.5% | Hurt |
| Vary the frequency and damping hyperparameters (ablation) | coRNN · Noise padded CIFAR-10 | test accuracy | No clear effect |
| Make timestep, frequency, and damping trainable rather than fixed hyperparameters (ablation) | coRNN · Learning tasks in the paper | performance | No clear effect |
| Use coRNN instead of LSTM for strongly chaotic dynamics | Time-series predictor · Lorenz 96 (F=8) | test NRMSE: $6.8\times 10^{-2}$ -> $9.8\times 10^{-2}$ | Hurt |
Closest papers by meaning
- UnICORNN: A recurrent model for learning very long time dependencies (ICML 2021 · similarity 0.91)
- Unconditional stability of a recurrent neural circuit implementing divisive normalization (NeurIPS 2024 · similarity 0.89)
- Recurrent neural networks: vanishing and exploding gradients are not the end of the story (NeurIPS 2024 · similarity 0.88)
- Learning Dynamics of RNNs in Closed-Loop Environments (NeurIPS 2025 · similarity 0.87)
- Recurrent neural network dynamical systems for biological vision (NeurIPS 2024 · similarity 0.87)
- Oscillatory State-Space Models (ICLR 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.