FCM Clustering from the View Point of Iteratively Reweighted Least Squares
H. Ichihashi, Katsuhiro Honda · 2005
By alleviating theoretical strictness, a broad class of membership functions can be used in fuzzy c-means (FCM) clustering from the viewpoint of iteratively reweighted least-squares (IRLS) techniques. Clustering characteristics of regular FCM, entropy regularized FCM or deterministic annealing by Rose and our proposed IRLS approaches are compared by using 3D graphics and contour maps. Though an in-depth analysis of theoretical aspect is beyond the scope of this paper, numerical comparisons reveal that IRLS algorithm using different membership functions do not share the same property with the regular FCM and entropy regularized FCM or DA