An Iterative Entropy Regularized Likelihood Learning Algorithm for Cluster Analysis with the Number of Clusters Automatically Detected

As for cluster analysis, the key problem is to determine the number of clusters. This paper presents an entropy regularized likelihood (ERL) learning principle for cluster analysis based on a mixture model to solve this problem. The well-known maximum likelihood estimation is just a special case of ERL learning. Moreover, when the regularization term is the same important as the log-likelihood, the ERL learning actually becomes Bayesian Ying-Yang (BYY) harmony learning. An iterative implementation is then proposed for ERL learning instead of gradient descent, the simulation and color image segmentation experiments have demonstrated that the iterative ERL learning algorithm can automatically detect the number of clusters during the parameter estimation