A new method for initializing radial basis function classifiers

Introduces a new approach for the selection of RBF kernel centers and their effective widths. RBF centers are divided into two sets and are placed strategically to maximize the classification capability of RBF networks. The first set is located near class boundaries at locations specified by a set of boundary-preserving patterns. The second set of RBF centers is represented by cluster centers using the k-means clustering algorithm. The widths of RBF kernels in both sets are selected so as to minimize the amount of overlap between different class regions. The merits of the authors' approach are validated using a speaker-independent vowel recognition problem.<<ETX>>

[1]  R.P. Lippmann,et al.  Pattern classification using neural networks , 1989, IEEE Communications Magazine.

[2]  Richard Lippmann,et al.  A Boundary Hunting Radial Basis Function Classifier which Allocates Centers Constructively , 1992, NIPS.

[3]  Mohamad T. Musavi,et al.  On the training of radial basis function classifiers , 1992, Neural Networks.

[4]  S. Dasgupta,et al.  Weight initialization of MLP classifiers using boundary-preserving patterns , 1994, Proceedings of 1994 IEEE International Conference on Neural Networks (ICNN'94).