增量学习的模糊风格K平面聚类  

Incremental learning based fuzzy style K-plane clustering

作　　者：顾苏杭 王士同[1] GU Su-hang;WANG Shi-tong(School of Digital Media,Jiangnan University,Wuxi 214122,China;School of Information Engineering and Technology,Changzhou Institute of Industry Technology,Changzhou 213164,China)

机构地区：[1]江南大学数字媒体学院,江苏无锡214122 [2]常州轻工职业技术学院信息工程与技术学院,江苏常州213164

出　　处：《控制与决策》2020年第9期2081-2093,共13页Control and Decision

基　　金：国家自然科学基金项目(61572236,61300151);常州工业职业技术学院博士基金项目(BSJJ13101010);常州工业职业技术学院新一代信息技术团队项目(YB201813101005)。

摘　　要：提出利用特征增量学习和数据风格信息双知识表达约束的模糊K平面聚类(ISF-KPC)算法.为了获得更好的泛化性,聚类前利用高斯核函数对原输入特征进行增长式的特征扩维.考虑数据集中来源于同一聚类的样本具有相同的风格,以矩阵的形式表达数据风格信息,并采用迭代的方式确定每个聚类的风格矩阵.大量实验结果表明,双知识表达约束的ISF-KPC与对比算法相比能够取得竞争性的聚类性能,尤其在具有典型风格数据集上能够取得优异的聚类性能.A fuzzy K-plane clustering algorithm based on double knowledge representations about incremental feature learning and homogeneous style of data(ISF-KPC) is proposed. Before partitioning data samples into different groups,i.e., clusters, the feature augumentation is firstly conducted in an incremental manner based on the original inputs. Since data samples originating from a group share a same homogeneous style, the style information of each group, denoted by a style matrix, will be iteratively determined. By extensive experiments, the proposed ISF-KPC based on the double knowledge representations can obtain comparative clustering performance when compared to the adopted comparative clustering methods, especially it has its best clustering performance on the datasets with clearly homogeneous styles.

关 键 词：K平面聚类 风格信息 特征扩维 模糊聚类 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]