Low-Rank Kernelized Graph-based Clustering using Multiple Views

Abstract

Kernelized methods using multiple kernels have shown better performances in graph-based clustering. But those kernelized methods get affected by the noise present in the data set. Also, only a single view has been used in those kernelized graph-based clustering methods. To address those issues, a novel low-rank multi-view multi-kernel graph-based clustering framework (LRMVMKC) has been proposed in this paper. Where the similarity nature of kernel matrices are exploited by low-rank optimal kernel learning and the clustering performances are boosted by using multiple views that provide different partial information about a given data set. The use of the proposed LRMVMKC framework on different benchmark data sets demonstrates the better performances of the proposed framework over other existing methods.