Computer Engineering and Applications ›› 2020, Vol. 56 ›› Issue (17): 60-68.DOI: 10.3778/j.issn.1002-8331.1907-0407

Previous Articles     Next Articles

Multi-view Subspace Clustering Network Based on Deep Autoencoder

GUO Sheng, ZHONG Zhaoman, LI Cunhua   

  1. 1.School of Computer Science and Technology, China University of Mining and Technology, Xuzhou, Jiangsu 221000, China
    2.School of Computer Engineering, Jiangsu Ocean University, Lianyungang, Jiangsu 222005, China
  • Online:2020-09-01 Published:2020-08-31

基于深度自编码的多视图子空间聚类网络

郭圣,仲兆满,李存华   

  1. 1.中国矿业大学 计算机科学与技术学院,江苏 徐州 221000
    2.江苏海洋大学 计算机工程学院,江苏 连云港 222005

Abstract:

Traditional subspace clustering model has a poor performance in multi-view data and nonlinear data. The paper is devoted to multi-view subspace clustering model based on deep autoencoder so as to improve the learning capability of multi-view space clustering algorithm by introducing the self-expression and weighted sparse representation of subspace clustering to the autoencoder. The target model can cluster data points with complex structure, which proves the effectiveness by multi-view dataset. The result shows that the method has a good performance in mining the multiple clustering structures of data effectively and complementing messages between multiple views. Compared with the existing methods, the performance of this method is greatly improved.

Key words: nonlinearity, deep autoencoder, self-representation, multi-view subspace clustering

摘要:

传统子空间浅层聚类模型对于多视图和非线性数据的聚类性能不佳。为此,提出一种基于深度自编码器的多视图子空间聚类网络模型,通过在深度自编码器中引入子空间聚类中的“自我表示”特性以及加权稀疏表示,提升了多视图子空间聚类算法的学习能力。推导的深度自编码多视图子空间聚类算法能够聚类具有复杂结构的数据点。通过多视图数据集验证了提出算法的有效性。结果表明,该方法能够有效地挖掘数据固有的多样性聚类结构,并利用多个视图之间互补信息,在性能上与现有方法相比有较大的提升。

关键词: 非线性, 深度自编码器, 自我表示, 多视图子空间聚类