On the role of context-specific independence in probabilistic inference

Context-specific independence (CSI) refers to conditional independencies that are true only in specific contexts. It has been found useful in various inference algorithms for Bayesian networks. This paper studies the role of CSI in general. We provide a characterization of the computational leverages offered by CSI without referring to particular inference algorithms. We identify the issues that need to be addressed in order to exploit the leverages and show how those issues can be addressed. We also provide empirical evidence that demonstrates the usefulness of CSI.