Title :
Training RBF networks with the Kalman filter
Author :
Ciocoiu, Iulian B.
Author_Institution :
Fac. of Electron. & Telecommun., Tech. Univ. of Iasi, Romania
Abstract :
We propose the use of Kalman filter as an alternative to the classical LMS algorithm for training Radial Basis Function (RBF) networks. A state-space formulation for the problem is given and the corresponding equations of the resulting training algorithm are obtained. Simulation results for an example of chaotic time series prediction are presented
Keywords :
Kalman filters; learning (artificial intelligence); neural nets; time series; Kalman filter; RBF networks; chaotic time series prediction; radial basis function networks; state-space formulation; training algorithm; Covariance matrix; Estimation theory; Filtering theory; Kalman filters; Least squares approximation; Multilayer perceptrons; Nonlinear equations; Predictive models; Radial basis function networks; State estimation;
Conference_Titel :
Circuits and Systems, 1998. ISCAS '98. Proceedings of the 1998 IEEE International Symposium on
Conference_Location :
Monterey, CA
Print_ISBN :
0-7803-4455-3
DOI :
10.1109/ISCAS.1998.703906
