Deterministic Annealing Framework in MMMs-Induced Fuzzy Co-Clustering and Its Applicability

Initialization problem is a significant issue in FCM-type clustering models, in which alternative optimization is often started with random initial partitions and can be trapped into local optima caused by bad initialization. The deterministic clustering approach is a practical procedure for utilizing a robust feature of very fuzzy partitions and tries to converge the iterative FCM process to a plausible solution by gradually decreasing the fuzziness degree. In this paper, the initialization sensitivity issue is considered in multinomial mixture models-induced fuzzy coclustering context and a new approach for implementing the deterministic annealing mechanism to fuzzy co-clustering is proposed. The advantages of the proposed approach against the conventional statistical co-clustering model are demonstrated through some numerical experiments.