Efficient non-convex graph clustering for big data

Big data analysis is a fundamental research topic with extensive technical obstacles yet to be overcome. Graph clustering has shown promise in addressing big data challenges by categorizing otherwise unlabeled data-thus giving them meaning. In this paper, we propose a set of non-convex programs, generally referred to as Hard and Soft Clustering programs, that rely on matrix factorization formulations for enhanced computational performance. Based on such formulations, we devise clustering algorithms that allow for large data analysis in a more efficient manner than traditional convex clustering techniques. Numerical results confirm the usefulness of the proposed algorithms for clustering purposes and reveal their potential for usage in big data applications.