Deep matrix factorization with knowledge transfer for lifelong clustering and semi-supervised clustering

Abstract Clustering analysis aims to group unlabeled data in an unsupervised learning manner. However, most existing methods are tailored for single-task data and do not work for a sequence of tasks. In this paper, we propose a Deep Matrix factorization method with Knowledge transfer (DMK) to address clustering problem in a lifelong setting, where DMK approaches a sequence of tasks; after each task is learned, its knowledge will be retained and later used to help future clustering task . To this end, we delve into deep matrix factorization and graph co-clustering, where (1) the former learns a basis feature library across all arrived tasks and a specific representation for each target task to deal with lifelong clustering and (2) the latter builds a consistent feature embedding library to transfer knowledge between each pair of tasks. An iterative optimization algorithm is then proposed to alternatively update the two libraries. In addition, we extend our DMK into a semi-supervised version and propose a Semi-supervised Deep Matrix factorization method with Knowledge transfer (SDMK) by exploiting a few of prior label information for lifelong semi-supervised clustering. Experimental results using four datasets with sequential tasks demonstrate that the proposed methods outperform state-of-the-art baseline methods markedly.