NEURAL NETWORKS FOR MAXIMUM-LIKELIHOOD CLUSTERING

Winner-take-all algorithms are commonly used techniques in clustering analysis. However, they have some problems ranging from clusters under utilization to the extended training time. Some solutions to these pr oblems are addressed here. It is shown here that using the maximum-lik elihood criterion instead of the Euclidean distance metric results in better clustering. The clusters are represented by a set of neuron eac h has a Gaussian receptive field. For these Gaussian neurons, the cova riance matrices, in addition to the centers, are learned. The one-winn er condition is relaxed by maximizing the likelihood function of the m ixture density function of the samples. This produces larger likelihoo d values and more normally distributed clusters. A fast mixture likeli hood clustering is provided for both batch and pattern learning modes. Convergence analysis and experimental results are also presented.