Learning Nonlinear Dynamical Systems Using the Expectation–Maximization Algorithm
Sam T. Roweis, Zoubin Ghahramani · 2001
This chapter addresses the problem of learning time-series models when the internal state is hidden. A brief review of the two fundamental algorithms that form the basis of the learning procedure is included. The algorithm and derivation of its learning rules is introduced. Results of using the algorithm to identify nonlinear dynamical systems is presented, as well as conclusions and potential extensions to the algorithm.