Journal of Frontiers of Computer Science and Technology ›› 2020, Vol. 14 ›› Issue (11): 1899-1907.DOI: 10.3778/j.issn.1673-9418.2002041

Previous Articles     Next Articles

Discriminative and Graph Regularized Nonnegative Matrix Factorization with Kernel Method

LI Xiangli, ZHANG Ying   

  1. 1. School of Mathematics and Computational Science, Guilin University of Electronic Technology, Guilin, Guangxi 541004, China
    2. Guangxi Key Laboratory of Cryptography and Information Security, Guilin, Guangxi 541004, China
    3. Guangxi Key Laboratory of Automatic Testing Technology and Instruments, Guilin, Guangxi 541004, China
    4. Guangxi University Key Laboratory of Data Analysis and Calculation, Guilin, Guangxi 541004, China
  • Online:2020-11-01 Published:2020-11-09

带核方法的判别图正则非负矩阵分解

李向利张颖   

  1. 1. 桂林电子科技大学 数学与计算科学学院,广西 桂林 541004
    2. 广西密码学与信息安全重点实验室,广西 桂林 541004
    3. 广西自动检测技术与仪器重点实验室,广西 桂林 541004
    4. 广西高校数据分析与计算重点实验室,广西 桂林 541004

Abstract:

Nonnegative matrix factorization (NMF) is a popular technique for dimension reduction,which has been extensively applied in image clustering and other fields.However,NMF is an unsupervised approach,which does not take the label information of the data and capture the inherent geometrical structure of data space.And NMF is a linear method that can't be used when the data are nonlinear.To this end,discriminative and graph regularized non-negative matrix factorization with kernel method is proposed,which uses the available label information,incorporates the graph into the NMF to capture the inherent geometrical structure and uses the kernel method to avoid the nonlinear data, and the result of factorization can effectively improve the clustering effect.Iterative initialization of variants of the NMF is random.A “warm start”strategy is adopted to avoid randomness in the result.Clustering experi-ments on several image datasets verify the effectiveness of the algorithm proposed in this paper compared with the other state-of-the-art methods.

Key words: nonnegative matrix factorization (NMF), semi-supervised clustering, graph regular, kernel method

摘要:

非负矩阵分解(NMF)是一种非常有效的数据降维方法,广泛应用于图像聚类等领域。然而NMF是一种无监督的方法,没有使用数据的标签信息,也不能捕获数据固有的几何结构,并且这是一种线性的方法,不能处理数据是非线性的情况。为此,提出了一种带核方法的判别图正则非负矩阵分解算法。该算法使用了部分有标签数据的标签信息,加入了图正则项来捕获数据的几何结构,使用核方法解决了数据非线性的问题,分解的结果能够有效地提高聚类效果。一般的非负矩阵分解迭代更新的初始化是随机产生的,使用一种“热启动”的策略,减小了结果的随机性。在几种图片数据集上使用该算法进行聚类实验,并与一些先进算法进行了比较,实验结果证明了该算法的有效性。

关键词: 非负矩阵分解(NMF), 半监督聚类, 图正则, 核方法