Semisupervised Fuzzy Clustering With Fuzzy Pairwise Constraints

In semisupervised fuzzy clustering, this article extends the traditional pairwise constraint (i.e., must-link or cannot-link) to constraint. The allows a supervisor provide grade of similarity dissimilarity between implicit vectors pair samples. This can represent more complicated relationship samples and avoid eliminating characteristics. Then, we propose clustering with constraints (SSFPC). nonconvex optimization problem in our SSFPC is solved by modified expectation-maximization algorithm, involving solve several indefinite quadratic programming problems (IQPPs). Further, diagonal block coordinate decent (DBCD) algorithm proposed for these IQPPs, whose stationary points are guaranteed, global solutions be obtained under certain conditions. To suit different applications, extended into various metric spaces, e.g., reproducing kernel Hilbert space. Experimental results on benchmark datasets facial expression database demonstrate outperformance compared some state-of-the-art models