Abstract

In this paper, we propose a general framework to accelerate significantly thealgorithms for nonnegative matrix factorization (NMF). This framework isinspired from the extrapolation scheme used to accelerate gradient methods inconvex optimization and from the method of parallel tangents. However, the useof extrapolation in the context of the two-block coordinate descent algorithmstackling the non-convex NMF problems is novel. We illustrate the performance ofthis approach on two state-of-the-art NMF algorithms, namely, acceleratedhierarchical alternating least squares (A-HALS) and alternating nonnegativeleast squares (ANLS), using synthetic, image and document data sets.

 

Quick Read (beta)