Constructing Flexible Dynamic Belief Networks From First-Order Probabilistic Knowledge Bases

The language of first-order logic, though successfully used in many applications, is not powerful enough to deal with uncertain knowledge. Probability theory alone is not a solution to this problem because it lacks the ability to quantify over a potentially infinite domain and the ability to discuss individuals, their properties, and the connections between them. First-order probabilistic logic [13] provides a framework that incorporates the advantages of both these representation languages. In this paper, we introduce a representation language that is a subset of the language of firstorder probabilistic logic. We show that this language is expressive enough to model complex domains. In particular, we allow reasoning over time and we allow procedural representations of conditional probability tables. To illustrate this, we give two knowledge bases, one modeling the propagation of infectious diseases and the other modeling the behavior of traffic. We provide an algorithm for our sublang..