Stanford 50: State of the Art and Future Directions of Computational Mathematics and Numerical Computing


  • March 30, 2007
  • 12:15 pm - 12:40 pm

An SVD-based approach to nonnegative matrix factorization

Steve Vavasis (University of Waterloo)

Nonnegative matrix factorization (NNMF) was introduced as a tool for datamining by Lee and Seung in 1999. NNMF attempts to approximate a matrix with nonnegative entries by a product of two low-rank matrices, also with nonnegative entries. We propose an approach for computing a NNMF that is based on an algorithm for singular value decomposition. Preliminary computational tests indicate that this method is able to identify features successfully in realistic datasets.

Parts of this talk represent joint work with Ali Ghodsi of University of Waterloo.

Stanford University Home Page