A Fast Start Base on Lanczos Algorithm for Symmetric Nonnegative Positive Definition Matrix Factorization

Abstract

Symmetric nonnegative matrix factorization(SNMF) has widely employed in many areas of applications. Symmetric nonnegative positive definition matrix factorization(SNPDMF) is a sub problem of SNMF. Authors proposed a Lanczos-based initialization method for SNPDMF, which can be combined with existing SNPDMF algorithms and achieve higher efficiency. Experiments shows that the SNPDMF algorithm which combined the proposed initialization method can converges to a better solution.