Maximum-entropy clustering algorithm and its global convergence analysis

摘要

Constructing a batch of differentiable entropy functions touniformly approximate an objec- tive function by means of themaximum-entropy principle, a new clustering algorithm, calledmaximum- entropy clustering algorithm, is proposed based onoptimization theory. This algorithm is a soft gener- alization of thehard C-means algorithm and possesses global convergence. Itsrelations with other clustering algorithms are discussed.