Incremental Multi-view Clustering using Barycentric Coordinate Representation

Unsupervised learning, particularly clustering, involves systematically categorizing objects based on inherent similarities. As a popular research field of clustering, multi-view data analysis enriches clustering by considering diverse perspectives. Persistent challenges in dynamic multi-view clustering focus on optimizing speed and maintaining quality with an increasing number of views. Motivated by the success of recent researches, we propose in this article an advanced framework of incremental multi-view clustering based on barycentric coordinates representation (IMBC). This framework refines object representation of each view and saves only the first m views as base kernels, then learns a consensus kernel. Updated m base kernels can be obtained via the combination of previous m base kernels and the representation of a new view. The proposed approach also offers the feasibility of incremental multi-modal clustering: its versatility extends to various data types and modalities, enhancing its applicability in diverse scenarios. Through the implementation of various standard clustering algorithms on the consensus kernel, our proposed approaches demonstrate efficient computational processing, achieving comparable clustering performance to existing approaches while incurring lower computational time.