UNSUPERVISED PATTERN-CLASSIFICATION BY NEURAL NETWORKS

A neural network is applied to the unsupervised pattern classification approach. Given a set of data consisting of unlabeled samples from se veral classes, the task of unsupervised classification is to label eve ry sample in the same class by the same symbol such that the data set is divided into several clusters. We consider the hypothesis that the data set is drawn from a finite mixture of Gaussian distributions. The network architecture is a two-layer feedforward type: the units of th e first layer are Gaussians and each correspond to one component of th e mixture. The output layer provides the probability density estimatio n of the mixture. The weighted competitive learning is used to estimat e the mean vectors and the non-diagonal covariance matrices of the Gau ssian units. The number of Gaussian units in the hidden layer is optim ized by informational criteria. Some of the results are reported, and the performance of this approach is evaluated.