Theory of learning of high-dimensional controlled non-linear dynamical systems (I): models and methods