An Adaptive Reasoning Approach for Ordering Multiple-Variable Hypotheses

Identifying a list of the most likely multiple-variable hypotheses in a Bayesian network is an important type of query commonly encountered in many problem domains. Yet, it received little attention in the past due to, at least in part, the limited success in dealing with the complexity problem of ordering exponential number of hypotheses. The objective of this research is to develop an eecient reasoning scheme for the derivation of a list of the most likely multiple-variable hypotheses. Various probabilistic properties will be explored in the development of such a reasoning scheme, so that current existing eecient algorithms for single-variable hypotheses can be extended to cope with the partial ordering of multiple-variable hypotheses. The complexity issue is discussed and also several examples are used to illustrate the eeectiveness of the reasoning scheme.