Parallel multiple nonnegative matrices factorization using graphics processing unit

Multiple Nonnegative Matrices Factorization (MNMF) is a promising method to study and analyze a dataset which has different types of features or relationships. However, due to the high computational cost, MNMF cannot meet the needs of time response for large-scale datasets. In this paper, we introduce a Parallel Multiple Nonnegative Matrices Factorization (PMNMF) approach which is implemented on Graphics Processing Unit (GPU) under the Compute Unified Device Architecture (CUDA) framework. Experimental studies demonstrate that PMNMF approach using GPU is able to obtain 100× speedup in comparison to the traditional multiple nonnegative matrices factorization under our experimental condition.