Belief Logic Programming with Cyclic Dependencies

Our previous work [26] introduced Belief Logic Programming (BLP), a novel form of quantitative logic programming with correlation of evidence. Unlike other quantitative approaches to logic programming, this new theory is able to provide accurate conclusions in the presence of uncertainty when the sources of information are not independent. However, the semantics defined in [26] is not sufficiently general—it does not allow cyclic dependencies among beliefs, which is a serious limitation of expressive power. This paper extends the semantics of BLP to allow cyclic dependencies. We show that the new semantics is backward compatible with the semantics for acyclic BLP and has the expected properties. The results are illustrated with examples of inference in a simple diagnostic expert system.