Title: Acceleration of computation for fuzzy c-means clustering
Authors: Takatsugu Yoshioka - Okayama University of Science (Japan)
Masahiro Kuroda - Okayama University of Science (Japan)
Yuichi Mori - Okayama University of Science (Japan) [presenting]
Abstract: Fuzzy c-means clustering (FCM), which is a nonhierarchical and soft clustering method, sometimes requires high computational cost due to the iterative convergence in the computation. To reduce the cost, a general procedure to accelerate the iterative computation such as alternating least squares has been proposed. This procedure generates a new accelerated convergent sequence using the vector epsilon algorithm based on the original convergent sequence in estimating two or more parameters alternatingly. Since the procedure can be applied to any computation which generates a linearly convergent sequence, it is applied to FCM, in which the membership matrix and the cluster centroid matrix are estimated alternatingly until convergence, to obtain the computational results faster than the original computation. Some numerical experiments demonstrate that the vector epsilon accelerated FCM accelerates the computation twice faster or more than the original one.