Soft Subspace Clustering With Entropy Constraints
Man Li, Lihong Wang · 2020
This paper investigates a class of soft subspace clustering algorithms which integrates the negative entropy term in the objective functions. It is necessary and hard to determine the coefficient of negative entropy in practice, which prevents the algorithms from applications. In order to solve the problem of parameter selection, a modified objective function with entropy constraints is proposed by moving the negative entropy from the objective function of ERKM (Entropy Regularization K-Means) algorithm to the constraints. The updating rules are given by theoretical analysis and the performance is evaluated experimentally using ten UCI datasets. The experimental studies demonstrate that the results of the proposed algorithm (ERKM+) outperform the original ERKM and other two k-means-type clustering algorithms in most cases.