Manfred Jaeger

Department of Computer Science, Aalborg University, Denmark.

Time: Wednesday 21.4.2004, 14:30
Place: Room A4-106, Fr. Bajersvej 7

First-order Probabilistic Modeling

Probabilistic reasoning has been an active field of research in Artificial Intelligence for about 25 years. For much of this time the focus was on what might be termed propositional probabilistic models: probability distributions on propositional interpretations. More recently, numerous efforts have been made to also develop semantic foundations and representation languages for richer types of models that also incorporate reasoning capabilities more closely related to predicate logic.

The "relational Bayesian network" language represents one approach to first-order probabilistic reasoning. In this talk I will introduce syntax and semantics of this language. It will be shown how relational Bayesian networks incorporate such classical types of probabilistic models as Markov chains and random graphs. Several examples will illustrate the broad applicability of this language also to modeling problems outside the core domain of Artificial Intelligence. Finally, I will show how relational Bayesian networks provide a new approach to convergence laws in finite model theory.