arxivstat.MLcs.LG2026-07-24
Learning Ergodic Dynamical Systems from a Finite Trajectory
Oleksii Kachaiev, Silvia Villa, Lorenzo Rosasco
We consider the problem of learning from a single finite trajectory of an ergodic stochastic dynamical system. More precisely, we study discrete-time autonomous stochastic systems defining time-homogeneous Markov processes. We first focus on estimating the optimal one-step predic…