Algorithms, Initializations, and Convergence for the Nonnegative Matrix Factorization

It is well known that good initializations can improve the speed and accuracy\nof the solutions of many nonnegative matrix factorization (NMF) algorithms.\nMany NMF algorithms are sensitive with respect to the initialization of W or H\nor both. This is especially true of algorithms of the alternating least squares\n(ALS) type, including the two new ALS algorithms that we present in this paper.\nWe compare the results of six initialization procedures (two standard and four\nnew) on our ALS algorithms. Lastly, we discuss the practical issue of choosing\nan appropriate convergence criterion.\n