Probabilistic Mercer Kernel Clusters
Zheng Rong Yang · 2006
Cluster analysis is one of the most important areas in machine learning. Most clustering algorithms are working in the Euclidean space, where the basic requisite is that each cluster has a hyper-ellipsoidal distribution. It has been recognized that this restriction may not be satisfied in many applications and some new ideas have been proposed [26], [17]. This paper investigates the construction of probabilistic Mercer kernel clusters using the maximum likelihood training procedure.