Research map / Architectures and efficiency / RNNs and normalization
A recurrent neural network without chaos
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 |
|---|---|---|---|
| replace a 5M-parameter LSTM with a 5M-parameter CFN | Recurrent language model · Penn Treebank test (no dropout) | perplexity: 105.1 -> 106.3 | Hurt |
| replace a 46.4M-parameter LSTM with a 46.4M-parameter CFN | Recurrent language model · Text8 development (no dropout) | perplexity: 139.9 -> 142.0 | Hurt |
| replace a 20M-parameter LSTM with a 20M-parameter CFN | Recurrent language model · Penn Treebank test (with dropout) | perplexity: 74.3 -> 74.9 | Hurt |
| replace a 50M-parameter LSTM with a 50M-parameter CFN | Recurrent language model · Penn Treebank test (with dropout) | perplexity: 71.8 -> 72.2 | Hurt |
| use 2 layers instead of 1 (ablation) | LSTM · Penn Treebank test (no dropout) | perplexity: 105.1 -> 110.6 | Hurt |
| use 1 layer instead of 2 (ablation) | LSTM · Text8 development (no dropout) | perplexity: 139.9 -> 140.8 | Hurt |
Closest papers by meaning
- Hierarchically Gated Recurrent Neural Network for Sequence Modeling (NeurIPS 2023 · similarity 0.89)
- Dynamical Isometry and a Mean Field Theory of RNNs: Gating Enables Signal Propagation in Recurrent Neural Networks (ICML 2018 · similarity 0.88)
- Fast-Slow Recurrent Neural Networks (NeurIPS 2017 · similarity 0.88)
- Efficient recurrent architectures through activity sparsity and sparse back-propagation through time (ICLR 2023 · similarity 0.87)
- M$^2$RNN: Non-Linear RNNs with Matrix-Valued States for Scalable Language Modeling (NeurIPS 2026 · similarity 0.86)
- Recurrent neural networks: vanishing and exploding gradients are not the end of the story (NeurIPS 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.