Parallel Nonnegative Matrix Factorization Algorithm on the Distributed Memory Platform

Nonnegative matrix factorization provides a new sight into the observed signals and has been extensively applied in face recognition, text mining and spectral data analysis. Despite the success, it is inefficient for the large-scale data set, due to the notoriously slow convergence of the multiplicative updating method. In this paper, we try to solve the problem through the parallel computing technique. Considering the limitation of the shared memory platform, the parallel algorithms are implemented on the distributed memory platform with the message passing interface library. Moreover, we adopt the two-layer cascade factorization strategy to eliminate the network consumption. The parallel implementations are evaluated on a 16-node Beowulf cluster with two data sets in different scale. The experiments demonstrate that the proposed method is effective in both precision and efficiency.