基于最大化间隔准则和成对约束的鲁棒半监督聚类方法

In order to overcome the drawback of the existent semi-supervised maximum margin clustering algorithm, i.e., it is insensitive to the violation for the pairwise constraints when there are a few samples that are very similar but from two different ground-truth classes, this paper proposes a new semi-supervised maximum margin clustering method. Firstly, a set of robust loss functions for violating the pairwise constraints is put forward based on the maximum margin principle, which is still able to effectively detect and penalize the violation for the pairwise constraints in the aforementioned difficult case. Secondly, an iterative algorithm is derived based on the constrained concave-convex procedure (CCCP) to solve the resulting optimization problem. The Experimental results have demonstrated the effectiveness of the proposed method.