This two-page extended abstract was accepted to be presented at the SIAM Workshop on Combinatorial Scientific Computing in October of 2009. The abstract is brief and incomplete, but shows the potential for non-negative matrix factorization to allow recovery of arbitrary mixture models when conventional algorithms are cleverly initialized. This research was not presented, however, due to the commencement of my employment at Adtuitive Inc., starting on October 3rd, in New York City (which company was subsequently acquired by Etsy Inc.).
Download: Non-Parametric Discrete Mixture Model Recovery via NNMF