System Identification from multiple trajectories

Leo Simpson

University of Freiburg and Tool-Temp AG

Tuesday, July 02, 2024, 11:00 - 11:59

Building 102 - SR 01-012

While the Maximum Likelihood Estimation method is one of the most performant method for parameter estimation of a stochastic dynamical system, it provides consistent estimates only in the case of infinitely long trajectories, (as opposed to infinitely many short trajectories).
On the other hand, Subspace Identification Methods learn a linear system from a set of short trajectories.
In this talk, we compare these two methods, and propose a System Identification procedure that combine ideas from these two.