NeurIPS paper: predicting dynamics beyond tipping points without knowing control parameters

DangerousFunny1371 · reddit · 2026-10-02

The authors present their NeurIPS 2026 paper "Topological Out-of-Domain Generalization in Dynamical Systems Reconstruction" (arXiv:2606.22969), tackling the hardest form of generalization in dynamical systems reconstruction (DSR) and time series forecasting: topological OOD generalization (OODG), where the dynamical regime itself changes (e.g., cyclic to chaotic).

Original post →

More from Research

Research channel →