Researchers have developed a novel, minimal architecture called DynaBase for zero-shot reconstruction of dynamical systems. This architecture, featuring a single-parameter piecewise affine map and a context selector, can reproduce various dynamical regimes and surprisingly outperforms existing models in both long-term statistics and short-term predictions. Additionally, a separate paper addresses topological out-of-domain generalization in dynamical systems reconstruction, focusing on predicting regime changes like bifurcations without explicit knowledge of control parameters. AI
IMPACT These advancements could lead to more robust and interpretable models for time series forecasting and scientific modeling.
RANK_REASON The cluster contains two research papers detailing novel architectures and methods for dynamical systems reconstruction.
- Dynamical systems reconstruction
- Neural ODEs
- NeurIPS2026
- PLRNNs
- Time series forecasting
- Topological Out-of-Domain Generalization in Dynamical Systems Reconstruction
- DynaBase
- Dynamical Systems
- Minimal Interpretable Architecture
- Zero-Shot Reconstruction
AI-generated summary · Google Gemini · from 2 sources. How we write summaries →