Learning the Informative Components in Nonnegative Matrix Factorization

In order to exploit the informative components hidden in nonnegative matrix factorization, an information theoretic learning method, termed ITNMF, is presented. Different from the existing NMF methods, the proposed method is able to handle the general objective optimization, and takes the conjugate gradient technique to enhance the iterative optimization. To tackle the null matrix factorization problem, the line search approach adopts the insured conditions while keeping the feasible step descendent. In addition, the function value based stopping rule is employed to achieve optimized efficiency. Experiments of pattern classification on the data sets under variant pose and illumination conditions reveal that the proposed method can outperform the existing methods.