Convex-Nonnegative Matrix Factorization with structure constraints

Abstract

Nonnegative Matrix Factorization (NMF) is of great use in finding basis information of non-negative data. In this paper, a novel Convex-NMF (CNMF) method is presented, called Structure Constrained Convex-Nonnegative Matrix Factorization (SCNMF). The idea of SCNMF is to extend the original Convex-NMF by incorporating the structure constraints into the Convex-NMF decomposition. The SCNMF seeks to extract the representation space that preserves the geometry structure. Finally, our experiment results are presented.