Training Rbf Networks With An Extended Kalman Filter Optimized Using Fuzzy Logic

Jun Wang,Li Zhu,Zhihua Cai,Wenyin Gong,Xinwei Lu

DOI: https://doi.org/10.1007/978-0-387-44641-7_33

2006-01-01

Abstract:In this paper we propose a novel training algorithm for RBF networks that is based on extended kalman filter and fuzzy logic. After the user choose how many prototypes to include in the network. the extended kalman filter simultaneously solves for the prototype vectors and the weight matrix. The fuzzy logic is used to cope with the devergence problem caused by the insufficiently known a priori filter statistics. Results are presented on RBF networks as applied to the Iris classification problem. It is shown that the use of the extended Kalman filter and fuzzy logic results in faster learning and better results than conventional RBF networks.