Clustering of noised and heterogeneous multi-view data with graph learning and projection decomposition

作者：

Highlights：

• A novel matrix decomposition strategy for multi-view data is proposed, where the consistency and diversity of various modals are simultaneously learned and explicitly quantified.

• CDP2D is a joint learning framework for MVC issue, which integrates selfrepresentation learning, projection decomposition, and affinity network construction.

• Experimental results indicate that CDP2D outperforms state-of-the-art methods in terms of accuracy. Moreover, CDP2D is a unified MVC framework which can handle both noise and incomplete multi-modal data.

摘要

•A novel matrix decomposition strategy for multi-view data is proposed, where the consistency and diversity of various modals are simultaneously learned and explicitly quantified.•CDP2D is a joint learning framework for MVC issue, which integrates selfrepresentation learning, projection decomposition, and affinity network construction.•Experimental results indicate that CDP2D outperforms state-of-the-art methods in terms of accuracy. Moreover, CDP2D is a unified MVC framework which can handle both noise and incomplete multi-modal data.

论文关键词：Multi-view clustering,Graph learning,Projection decomposition,Matrix factorization

论文评审过程：Received 16 June 2022, Revised 12 August 2022, Accepted 16 August 2022, Available online 23 August 2022, Version of Record 5 September 2022.

论文官网地址：https://doi.org/10.1016/j.knosys.2022.109736