A rival penalized EM algorithm towards maximizing weighted likelihood for density mixture clustering with automatic model selection

摘要

How to determine the number of clusters is the intractable problem in clustering analysis. We propose a new learning paradigm named maximum weighted likelihood (MwL), in which the weights can be designed. Accordingly, we develop a novel rival penalized expectation-maximization (RPEM) algorithm, whose intrinsic rival penalization mechanism enables the redundant densities in the mixture to be gradually faded out during the learning. Hence, the RPEM can automatically select an appropriate number of densities in the density mixture clustering. The experiments have shown promising results.