Research Seminars

Structured Non-Negative Matrix Factorization

Hans Laurberg
Aalborg University, Denmark

Friday 8 June 2007, 4:00pm, Room 105

Abstract

Non-negative matrix factorization (NMF) has been widely used since the initial paper of Lee and Seung. In this talk, NMF will be introduced and compared with principal component analysis. For particular data sets, most NMF algorithms fail to resolve the underlying structure and it is shown that these data sets exhibit the same characteristics. An efficient structured NMF algorithm that can resolve this structure by assuming several sources is proposed. Finally, it is shown how the assumption of several sources can be applied to musical instrument separation.