Author:

Keywords:

gaussian mixture model (gmm), radial basis function (rbf), expectation-maximization (em), maximum-likelihood (nil), Science & Technology, Technology, Computer Science, Artificial Intelligence, Computer Science, Theory & Methods, Computer Science, Gaussian mixture model (GMM), radial basis function (RBF), expectation-maximization (EM), maximum-likelihood (NIL)

Abstract:

In paper we propose a Bayesian classifier for multiclass problem by using the merging RBF networks. The estimation of probability density function (PDF) with a Gaussian mixture model is used to update the expectation maximization algorithm. The centers and variances of RBF networks are gradually updated to merge the basis unites by the supervised gradient descent of the error energy function. The algorithms are used to construct the RBF networks and to reduce the number of basis units. The experimental results show the validity of our method which gives a smaller number of basis units and obviously outperforms the conventional RBF learning technique.