Research map / Architectures and efficiency / RNNs and normalization
Legendre Memory Units: Continuous-Time Representation in 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 |
|---|---|---|---|
| LMU instead of NRU | RNN with 468 state variables · psMNIST test | accuracy: 95.38 -> 97.15 | Helped |
| LMU instead of LSTM | 4-layer ~18k param RNN · Mackey-Glass | NRMSE: 0.079 -> 0.054 | Helped |
| LMU instead of LSTM | 4-layer RNN · Mackey-Glass | training time per epoch: 20.34 s -> 12.89 s | Helped |
| interleave LSTM layers with LMU layers (hybrid) (ablation) | LMU · Mackey-Glass | NRMSE: 0.054 -> 0.050 | Helped |
| perturb diagonals of memory matrix A away from Pade/Legendre values (ablation) | LMU · psMNIST | test accuracy | Hurt |
| LMU memory cell instead of LSTM | capacity task · capacity task T>=400 | test MSE | Helped |
Closest papers by meaning
- Parallelizing Legendre Memory Unit Training (ICML 2021 · similarity 0.89)
- Recurrent neural networks: vanishing and exploding gradients are not the end of the story (NeurIPS 2024 · similarity 0.88)
- Understanding and Controlling Memory in Recurrent Neural Networks (ICML 2019 · similarity 0.88)
- UnICORNN: A recurrent model for learning very long time dependencies (ICML 2021 · similarity 0.88)
- Efficient recurrent architectures through activity sparsity and sparse back-propagation through time (ICLR 2023 · similarity 0.88)
- Exploring the Promise and Limits of Real-Time Recurrent Learning (ICLR 2024 · similarity 0.86)
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.