Fast Nonnegative Matrix Factorization Using Nested ADMM Iterations

Abstract

As a constrained low-rank decomposition technique nonnegative matrix factorization (NMF) finds a wide variety of applications, especially in the analysis and design of pattern recognition systems for large-scale datasets. In this paper, we present a new algorithm for NMF based on nested alternating direction method of multipliers (ADMM) iterations. The paper describes the algorithm with a great deal of technical details, and includes a case study to demonstrate the algorithm’s ability to handle large-scale datasets with improved efficiency in comparison with those not using nested ADMM iterations.