Bayesian inference algorithm on Raw

Abstract

This work explores the performance of Raw, a parallel hardware platform developed at MIT, running a Bayesian inference algorithm. Motivation for examining this parallel system is a growing interest in creating a self-learning and cognitive processor, which these hardware and software components can potentially produce. The Bayesian inference algorithm is mapped onto Raw in a variety of ways to try to account for the fact that different implementations give different processor performance. Results for the processor performance, determined by looking at a wide variety of metrics look promising, suggesting that Raw has the potential to successfully run such algorithms.