On-line EM algorithm and reconstruction of chaotic dynamics
Shin Ishii, Masa-aki Sato · 2002
We previously (1998) proposed an online EM algorithm for the normalized Gaussian network model, which is a network of local linear regression units. In this paper, we apply our approach to an identification problem of unknown nonlinear dynamics. Our approach is able to reconstruct the dynamics in shorter learning steps than approaches based on the recurrent neural network model. Even when dynamical variables can partially be observed, our approach is able to well reproduce the trajectory of the observed variables.