计算机工程与应用 ›› 2022, Vol. 58 ›› Issue (13): 154-163.DOI: 10.3778/j.issn.1002-8331.2106-0157

• 模式识别与人工智能 • 上一篇    下一篇

面向多视图聚类的低秩张量表示学习

余瑶,杜世强,宋金梅   

  1. 1.西北民族大学 数学与计算机科学学院,兰州 730030
    2.西北民族大学 中国民族信息技术研究院,兰州 730030
  • 出版日期:2022-07-01 发布日期:2022-07-01

Low-Rank Tensor Representation Learning for Multi-View Clustering

YU Yao, DU Shiqiang, SONG Jinmei   

  1. 1.School of Mathematics and Computer Science, Northwest Minzu University, Lanzhou 730030, China
    2.School of China National Institute of Information Technology, Northwest Minzu University, Lanzhou 730030, China
  • Online:2022-07-01 Published:2022-07-01

摘要: 针对现有鲁棒图学习忽略多视图间的互补信息和高阶相关性问题,提出一种面向多视图聚类的低秩张量表示学习(LRTRL-MVC)算法。利用鲁棒主成分分析的思想,在去除噪声的干净数据上计算各视图的鲁棒图和转移概率矩阵,然后构建一个包含各视图马尔可夫转移概率矩阵的张量,采用基于张量奇异值分解的核范数来确保目标张量的低秩性质。利用迭代最优化算法求解,将求得的低秩张量作为马尔可夫谱聚类算法的输入得到最终聚类结果。在4个不同类型的公开标准数据集BBCSport、NGs、Yale和MSRCv1上进行实验并与相关的最好多视图聚类算法进行对比,结果表明在3个聚类度量标准下,所提算法的聚类结果均高于其他对比算法。

关键词: 多视图聚类, 马尔可夫链, 低秩张量

Abstract: Aiming at the problem that the existing robust graph learning ignores the complementary information and high-order correlation between multiple views, a low-rank tensor representation learning for multi-view clustering(LRTRL-MVC) is proposed. Firstly, using the idea of robust principal component analysis, the robust graph and transition probability matrix of each view are calculated on the clean data with noise removed, and then a tensor containing the Markov transition probability matrix of each view is constructed. Finally, the nuclear norm based on tensor singular value decomposition is used to ensure the low rank property of the target tensor. The low rank tensor is used as the input of Markov spectral clustering algorithm to obtain the final clustering result. Experiments are carried out on four different types of public standard datasets BBCSport, NGs, Yale and MSRCv1, and compared with the related best multi view clustering algorithm. The results show that the clustering results of the proposed algorithm are higher than other comparison algorithms under the three clustering metrics.

Key words: multi-view clustering, Markov chain, low-rank tensor