FCM-type fuzzy co-clustering by K-L information regularization
Katsuhiro Honda, Shunnya Oshio, Akira Notsu · 2014
Fuzzy c-Means (FCM) clustering by entropy-based regularization concept is a fuzzy variant of Gaussian mixtures density estimation. FCM was also extended to a full-parameter model by introducing Mahalanobis distance and the K-L information-based fuzzification scheme, in which the degree of fuzziness of partition is evaluated comparing with Gaussian mixtures. In this paper, a new fuzzy co-clustering model is proposed, which is a fuzzy variant of multinomial mixture density estimation. Multinomial mixtures is a probabilistic model for co-clustering of cooccurrence matrices and the proposed method extends multinomial mixtures so that the degree of fuzziness can be tuned in a similar manner to K-L information-based FCM. Several experimental results demonstrate the effects of tuning the degree of fuzziness comparing with its corresponding probabilistic model.