Research map / Architectures and efficiency / RNNs and normalization
Stable Recurrent Models
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 |
|---|---|---|---|
| Constrain the recurrent weight matrix to have spectral norm below 1 | RNN · JSB Chorales | negative log-likelihood: 8.9 -> 8.9 | No clear effect |
| Constrain the recurrent weight matrix to have spectral norm below 1 | RNN · ATIS | F1 score: 94.7 -> 94.7 | No clear effect |
| Constrain the recurrent weight matrix to have spectral norm below 1 | RNN · Wikitext-2 | perplexity: 146.7 -> 143.5 | Helped |
| Constrain the recurrent weight matrix to have spectral norm below 1 | RNN · Penn Treebank | bits per character: 1.8 -> 1.9 | Hurt |
| Normalize weight-matrix rows to enforce the sufficient stability conditions | LSTM · JSB Chorales | negative log-likelihood: 8.5 -> 8.5 | No clear effect |
| Normalize weight-matrix rows to enforce the sufficient stability conditions | LSTM · ATIS | F1 score: 95.1 -> 94.6 | Hurt |
| Normalize weight-matrix rows to enforce the sufficient stability conditions | LSTM · Wikitext-2 | perplexity: 95.7 -> 113.2 | Hurt |
| Normalize weight-matrix rows to enforce the sufficient stability conditions | LSTM · Penn Treebank | bits per character: 1.4 -> 1.9 | Hurt |
Closest papers by meaning
- RNNs of RNNs: Recursive Construction of Stable Assemblies of Recurrent Neural Networks (NeurIPS 2022 · similarity 0.90)
- AntisymmetricRNN: A Dynamical System View on Recurrent Neural Networks (ICLR 2019 · similarity 0.90)
- Lipschitz Recurrent Neural Networks (ICLR 2021 · similarity 0.89)
- Recurrent neural networks: vanishing and exploding gradients are not the end of the story (NeurIPS 2024 · similarity 0.89)
- Learning dynamics in linear recurrent neural networks (ICML 2025 · similarity 0.88)
- Unconditional stability of a recurrent neural circuit implementing divisive normalization (NeurIPS 2024 · similarity 0.88)
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.