Sequential RBF function estimator: memory regression network

• Published in: 2004 IEEE International Conference on Systems, Man and Cybernetics (IEEE Cat. No.04CH37583) Vol. 5; pp. 4815 - 4820 vol.5
• Main Authors: CHOW, Chi-Kin, TSUI, Hung-Tat
• Format: Conference Proceeding
• Language: English
• Published: Piscataway NJ IEEE 2004
• Subjects: Applied sciences; Computer science; control theory; systems; Control theory. Systems; Exact sciences and technology; Humans; Image processing; Interpolation; Laboratories; Neural networks; Neurons; Radial basis function networks; Radio access networks; Robots; Signal processing algorithms; Short term; Radial basis function; Batch mode; Observer; Regression function; Set computation; Regression analysis; Neural network; Memory function; Long term; Human factor; Mean square error
• ISBN: 0780385667; 9780780385665
• ISSN: 1062-922X; 2577-1655
• DOI: 10.1109/ICSMC.2004.1401293

The neural-network training algorithm can be divided into 2 categories: (1) batch mode and (2) sequential mode. In this paper, a novel online RBF network called "memory regression network (MRN)" is proposed. Different from the previous approaches (de Freitas, N, et al., Aug. 1999), (Schiffman, W, et al., 1993), MRN involves two types of memories: experience and neuron, which handle short and long term memories respectively. By simulating human's learning behavior, a given function can be estimated without memorizing the whole training set. Two sets of function estimation experiments are examined in order to illustrate the performance of the proposed algorithm. The results show that MRN can effectively approximate the given function within a reasonable time and acceptable mean square error