Probabilistic Latent Variable Models as Non-Negative Factorizations

    •  Madhusudana Shashanka, Bhiksha Raj, Paris Smaragdis, "Probabilistic Latent Variable Models as Non-Negative Factorizations", Tech. Rep. TR2007-083, Mitsubishi Electric Research Laboratories, Cambridge, MA, December 2007.
      BibTeX TR2007-083 PDF
      • @techreport{MERL_TR2007-083,
      • author = {Madhusudana Shashanka, Bhiksha Raj, Paris Smaragdis},
      • title = {Probabilistic Latent Variable Models as Non-Negative Factorizations},
      • institution = {MERL - Mitsubishi Electric Research Laboratories},
      • address = {Cambridge, MA 02139},
      • number = {TR2007-083},
      • month = dec,
      • year = 2007,
      • url = {}
      • }
  • Research Areas:

    Artificial Intelligence, Speech & Audio


In this paper we present a family of probabilistic latent variable models which can be used for analysis of non-negative data. We show that there strong ties between non-negative matrix factorization and this family, and we also provide some straightforward extensions which can help in dealing with shift-invariances, higher order decompositions and sparsity constraints. Through these extensions we argue that the use of this approach allows for rapid development of complex statistical models for analyzing non-negative data.