Nonlinear state space learning with EM and neural networks

Author

De Freitas, Joiio Fg ; Niranjan, Mahesan ; Gee, Andrew H.

Author_Institution

Dept. of Eng., Cambridge Univ., UK

fYear

1998

fDate

31 Aug-2 Sep 1998

Firstpage

254

Lastpage

263

Abstract

In this paper, we derive the EM algorithm for nonlinear state space models. We show how this algorithm, in conjunction with the well known techniques of Kalman smoothing, can be used for nonlinear system identification. A multilayer perceptron, whose derivatives are computed by backpropagation, is used to generate the measurements mapping. We found that the methodic is intrinsically very powerful, simple, elegant and stable. However, it exhibits very slow convergence

Keywords

Kalman filters; backpropagation; convergence; identification; maximum likelihood estimation; multilayer perceptrons; nonlinear systems; smoothing methods; state-space methods; EM algorithm; Kalman smoothing; backpropagation; convergence; measurements mapping; multilayer perceptron; neural networks; nonlinear state space learning; nonlinear system identification; Covariance matrix; Hidden Markov models; Inference algorithms; Kalman filters; Neural networks; Power system modeling; Smoothing methods; State-space methods; Switches; Working environment noise;

fLanguage

English

Publisher

ieee

Conference_Titel

Neural Networks for Signal Processing VIII, 1998. Proceedings of the 1998 IEEE Signal Processing Society Workshop

Conference_Location

Cambridge

ISSN

1089-3555

Print_ISBN

0-7803-5060-X

Type

conf

DOI

10.1109/NNSP.1998.710655

Filename

710655