Robust Multiview Subspace Clustering with Consistency Graph Learning


  • Received:2021-06-21 Revised:2021-07-16 Published:2021-09-03

CCML2021+149: 基于一致图学习的鲁棒多视图子空间聚类


  1. 山东师范大学
  • 通讯作者: 潘振君

Abstract: Focused on the problem that multi-view data analysis was susceptible to the noise of the original data set and required additional steps to calculate the clustering results, a robust multi-view subspace (RMCGL) clustering algorithm based on consistent graph learning was proposed. Firstly, the potential robust representation of data in the subspace was learned in each view, and the similarity matrix of each view was obtained based on this representation. Then a unified similarity graph was learned based on the obtained multiple similarity matrices. Finally, by adding rank constraints to the Laplace matrix corresponding to the similarity graph, the obtained similarity graph had the optimal clustering structure, and the final clustering results could be obtained directly. The process was completed in a unified optimization framework, which could simultaneously learn potential robust representations, similarity matrices and consensus graphs. The clustering Accuracy (ACC) of the RMCGL algorithm is improved by 3.36%, 5.82%, and 5.71% on the BBC, 100leaves, and MSRC datasets, respectively, compared with the graph-based multi-view clustering algorithm(GMC). Experimental results show that the algorithm has a good clustering effect.

Key words: multi-view, subspace, clustering, self-weighted, graph learning

摘要: 针对多视图数据分析易受原始数据集噪声干扰,以及需要额外的步骤计算聚类结果的问题,提出了一种基于一致图学习的鲁棒多视图子空间(RMCGL)聚类算法。该算法首先在各个视图下学习数据在子空间中的潜在鲁棒表示,并基于该表示得到各视图的相似度矩阵。随后基于得到的多个相似度矩阵学习一个统一的相似性图。最后,通过对相似图对应的拉普拉斯矩阵添加秩约束,确保得到的相似图具有最优的聚类结构,并可直接得到最终的聚类结果。该过程在一个统一的优化框架中完成,能同时学习潜在鲁棒表示、相似性矩阵和一致图。RMCGL算法的聚类精度(ACC)在BBC、100leaves和MSRC数据集上比基于图的多视图聚类算法(GMC)分别提升了3.36%、5.82%和5.71%。实验结果表明,该算法具有良好的聚类效果。

关键词: 多视图, 子空间, 聚类, 自加权, 图学习

CLC Number: