Theoretical analysis and practical insights on importance sampling in Bayesian networks

The AIS-BN algorithm [2] is a successful importance sampling-based algorithm for Bayesian networks that relies on two heuristic methods to obtain an initial importance function: 2-cutoff , replacing small probabilities in the conditional probability tables by a larger 2, and setting the probability distributions of the parents of evidence nodes to uniform. However, why the simple heuristics are so effective was not well understood. In this paper, we point out that it is due to a practical requirement for the importance function, which says that a good importance function should possess thicker tails than the actual posterior probability distribution. By studying the basic assumptions behind importance sampling and the properties of importance sampling in Bayesian networks, we develop several theoretical insights into the desirability of thick tails for importance functions. These insights not only shed light on the success of the two heuristics of AIS-BN, but also provide a common theoretical basis for several other successful heuristic methods.