Research map / Architectures and efficiency / RNNs and normalization
UnICORNN: A recurrent model for learning very 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 UnICORNN instead of coRNN | RNNs · psMNIST test | test accuracy: 97.3% -> 98.4% | Helped |
| Use UnICORNN instead of coRNN | RNNs · Noise padded CIFAR-10 test | test accuracy: 59.0% -> 62.4% | Helped |
| Use UnICORNN instead of coRNN | RNNs · EigenWorms test | test accuracy: 86.7% $\pm$ 3.0\% -> 90.3% $\pm$ \textbf{3.0}\% | Helped |
| Use UnICORNN instead of coRNN | RNNs · Healthcare RR test | L2 test error: 1.45 $\pm$ 0.23 -> \textbf{1.06} $\pm$ \textbf{0.03} | Helped |
| Use UnICORNN instead of coRNN | RNNs · Healthcare HR test | L2 test error: 1.71 $\pm$ 0.1 -> \textbf{1.39} $\pm$ \textbf{0.09} | Helped |
| Use UnICORNN instead of coRNN | RNNs · IMDB test | test accuracy: 87.4 \% -> \textbf{88.4}\% | Helped |
| Vary the maximum allowed time step Δt (ablation) | UnICORNN · Noise padded CIFAR-10 | test accuracy: approximately 15\% | Mixed |
| Vary the control parameter α (ablation) | UnICORNN · Noise padded CIFAR-10 | test accuracy | No clear effect |
| Train using only the last sequence entries (ablation) | UnICORNN · EigenWorms | test accuracy | Hurt |
| Train using a random subset of sequence entries instead of only the last entries (ablation) | UnICORNN · EigenWorms | test accuracy | Helped |
| Increase the number of sampled sequence entries (ablation) | UnICORNN · EigenWorms | test accuracy | Helped |
Closest papers by meaning
- Coupled Oscillatory Recurrent Neural Network (coRNN): An accurate and (gradient) stable architecture for learning long time dependencies (ICLR 2021 · similarity 0.91)
- Recurrent neural networks: vanishing and exploding gradients are not the end of the story (NeurIPS 2024 · similarity 0.90)
- AntisymmetricRNN: A Dynamical System View on Recurrent Neural Networks (ICLR 2019 · similarity 0.90)
- RNNs Incrementally Evolving on an Equilibrium Manifold: A Panacea for Vanishing and Exploding Gradients? (ICLR 2020 · similarity 0.90)
- Learning dynamics in linear recurrent neural networks (ICML 2025 · similarity 0.87)
- Emergent mechanisms for long timescales depend on training curriculum and affect performance in memory tasks (ICLR 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.