基于交叉熵测度的成对约束模糊核聚类算法  

Cross entropy measurement based fuzzy kernel clustering algorithm with pairwise constraints

作　　者：徐圣兵[1,2] 林上钧 钟国祥 XU Shengbing;LIN Shangjun;ZHONG Guoxiang(School of Computers,Guangdong University of Technology,Guangzhou 510006,China;School of Applied Mathematics,Guangdong University of Technology,Guangzhou 510520,China)

机构地区：[1]广东工业大学计算机学院,广东广州510006 [2]广东工业大学应用数学学院,广东广州510520

出　　处：《应用科技》2020年第1期80-87,共8页Applied Science and Technology

基　　金：国家自然科学基金项目(61672169);广东大学生科技创新培育专项资金(pdjhb0163).

摘　　要：目前已有的成对约束模糊核聚类研究中,缺乏对成对约束信息的有效测度,进而无法充分利用成对约束这类半监督信息。在成对约束核聚类的基础上,文中提出基于交叉熵测度的成对约束核聚类算法。利用对象交叉熵测度工具,提出最小-最大交叉熵隶属度学习准则,并作为成对约束信息测度项引入到成对约束核聚类的目标函数中,通过拉格朗日最优化处理目标函数,推导出相应聚类算法。实验进一步表明,该算法能够更有效利用成对约束半监督信息提升聚类性能。In the existing studies of fuzzy kernel clustering with pairwise constraints,little work had been done on the effective measurement of pairwise constraints information,so the semi-supervised information of pairwise constraints could not be utilized fully.In this paper,we propose an algorithm of cross entropy measurement based fuzzy kernel clustering with pairwise constraints(CEM-FKCPC).Using the sample cross-entropy measurement tool,the minimummaximum cross entropy membership learning criterion is put forward,and further introduced into the objective function of CEM-FKCPC as a measurement item of pairwise constraints information.The corresponding CEM-FKCPC algorithm can be derived by processing the objective function with the Lagrange optimization procedure.Experiments further show that the algorithm can improve the clustering performance by making use of the semi-supervised information of pairwise constraint more effectively.

关 键 词：成对约束 交叉熵 半监督 核聚类 模糊 隶属度 学习准则 拉格朗日最优化 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]