Nonnegative/binary matrix factorization with a D-Wave quantum annealer

Single-core computational performance relentlessly improved for decades, but recently that progress has begun to slow [7]. As a result, alternative computational architectures have sprung up including multi-core processors [8], graphic processing units [9], neuromorphic computing [10], and application-specific integrated circuits to name a few. Here we explore the use of another new architecture: quantum annealing [11]. In particular, we utilize the form of quantum annealing realized with D-Wave hardware [1, 3]. We focus on a machine learning problem based on matrix factorizations, and describe an algorithm for computing these matrix factorizations that leverages D-Wave hardware. We apply the algorithm to learn features in a set of facial images. There is an ongoing back-and-forth regarding whether or not D-Wave's hardware provides performance benefits over classical single-core computing [12, 13, 14, 15].