Multiview clustering has attracted much attention due to its ability to aggregate various source information and many advanced approaches have been proposed in the literature. However, there are still two major issues that need to be further explored: i) how to efficiently handle large-scale data; ii) how to effectively incorporate the complementary multiple sources. In this article, we fulfill a unified multiview subspace clustering model termed anchor-based multiview subspace clustering with diversity regularization by seamlessly optimizing subspace learning and multiview fusion. First, we efficiently evaluate the self-expression similarity matrix based on sampling anchor points to reduce the high time complexities in former methods. A regularization term is further imposed to encourage high independence and diversity of each view. In addition, we theoretically analyze the time complexity of the proposed algorithm. Comprehensive experiments on several benchmark datasets demonstrate that our proposed model consistently outperforms over the state-of-the-art techniques.

中文翻译：

---

具有多样性正则化的基于锚的多视图子空间聚类

多视图聚类由于能够聚合各种源信息而备受关注，并且文献中提出了许多高级方法。但是，仍然有两个主要问题需要进一步探讨：i）如何有效处理大规模数据；ii）如何有效地整合互补的多种资源。在本文中，我们通过无缝优化子空间学习和多视图融合，实现了一个具有多样性正则化的统一多视图子空间聚类模型，该模型称为基于锚的多视图子空间聚类。首先，我们基于采样锚点来有效地评估自表达相似度矩阵，以减少以前方法中的高时间复杂度。进一步强加一个正规化术语以鼓励每个视图的高度独立性和多样性。此外，我们从理论上分析了所提出算法的时间复杂度。在几个基准数据集上进行的综合实验表明，我们提出的模型始终优于最新技术。