Nonnegative matrix factorization for pattern recognition

In this paper, linear and unsupervised dimensionality re-duction via matrix factorization with nonnegativity con-straints is studied when applied for feature extraction, fol-lowed by pattern recognition. Since typically matrix fac-torization is iteratively done, convergence can be slow. To alleviate this problem, a significantly (more than 11 times) faster algorithm is proposed, which does not cause severe degradations in classification accuracy when dimensional-ity reduction is followed by classification. Such results are due to two modifications of the previous algorithms: fea-ture scaling (normalization) prior to the beginning of iter-ations and initialization of iterations, combining two tech-niques for mapping unseen data. KEY WORDS pattern recognition, nonnegative matrix factorization