An Incremental local Radial Basis Function Network

摘要

In this paper we present a learning algorithm to approximate functions which are piece-wise continuous.rnThe network is a RBF net in which the normalized functions are restricted to finite domains, and their variances are updated by supervised training. The net architecture grows during the training phase, increasing the number of hidden nodes in order to use a finer resolution where it is needed. The ability of this kind of net to save training time depends on selectively growing the net structure and on the locality of the basis activation functions. Some examples of the net performance are shown.