Probabilistic Partial Evaluation: Exploiting Rule Structure in Probabilistic Inference

Bayesian belief networks have grown to prominence because they provide compact representations of many domains, and there are algorithms to exploit this compactness. The next step is to allow compact representations of the conditional probability tables of a variable given its parents. In this paper we present such a representation in terms of parent contexts and provide an algorithm that exploits this compactness. The representation is in terms of rules that provide conditional probabilities in different contexts. The algorithm is based on eliminating the variables not needed in an answer in turn. The operations for eliminating a variable correspond to a form of partial evaluation, where we are careful to maintain the probabilistic dependencies necessary for correct probabilistic inference. We show how this new method can exploit more structure than previous methods for structured belief network inference.