Research map / Architectures and efficiency / RNNs and normalization
On Scrambling Phenomena for Randomly Initialized Recurrent Networks
Vaggos Chatziafratis, Ioannis Panageas, Clayton Sanford, Stelios Andrew Stavroulakis
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (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 He initialization for the recurrent weights | RNN(k, σ²) · RNN(k, σ²) | probability of Li-Yorke chaos: at least δ | Improved |
| Reduce weight variance to 1/(4k log k) | RNN(k, σ²) · RNN(k, σ²) | probability of Li-Yorke chaos: at most 1/k | Hurt |
| Increase weight scale to σ = ω(1) | RNN(k, σ²) · RNN(k, σ²) | probability of Li-Yorke chaos: lim_{k → ∞} Pr{f_k is Li-Yorke chaotic} = 1 | Improved |
| Use weights with variance ω(1/k) | RNN(k, σ²) · RNN(k, σ²) | probability of period 3: lim_{k → ∞} Pr{f_k is 3-periodic} = 1 | Improved |
| Use He initialization | RNN(k, σ²) · RNN feedback iterations | number of linear regions: exponential in the number t of feedback iterations | Improved |
| Reduce weight variance to 1/(4k log k) | RNN(k, σ²) · RNN(k, σ²) | probability of having no fixed points: ≥ 1-1/k | Improved |
Closest papers by meaning
- Recurrent neural networks: vanishing and exploding gradients are not the end of the story (NeurIPS 2024 · similarity 0.86)
- Slow Transition to Low-Dimensional Chaos in Heavy-Tailed Recurrent Neural Networks (NeurIPS 2025 · similarity 0.86)
- Mechanistic Interpretability of RNNs emulating Hidden Markov Models (NeurIPS 2025 · similarity 0.86)
- When are dynamical systems learned from time series data statistically accurate? (NeurIPS 2024 · similarity 0.86)
- Dynamical phases of short-term memory mechanisms in RNNs (ICML 2025 · similarity 0.86)
- Uniform-in-Time Weak Propagation of Chaos in Shallow Neural Networks (NeurIPS 2026 · 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.
