Maximum-entropy learning algorithm of radial basis function (RBF) neural networks

The key point in the design of RBF networks is to specify the number and the location of the centers. Several heuristic hybrid learning methods, which apply a clustering algorithm for locating the centers and subsequently a linear least squares method for the linear weights, have been previously suggested. A maximum-entropy clustering method for training the center vectors is constructed via the maximum-entropy principle in the information theory. Accordingly, a maximum-entropy learning algorithm (MELA) of the RBF networks is given. Two experiments, including time series prediction and system identification, are given to test MELA. The results show, compared with the error back-propagating algorithm of the multi-layer perceptions (MLP) and RBF, MELA not only improves learning precision and generalization ability, but reduces learning time as well.