Averaged Probabilistic Relational Models

Most real-world data is stored in relational form. In contrast, most statistical learning methods work with “flat” data representations, forcing us to convert our data into a form that loses much of the relational structure. The recently introduced framework of Probabilistic Relational Models (PRMs) allows us to represent probabilistic models over multiple entities that utilize the relations between them. However, for extremely large domains it may be impossible to represent every object and every relation in the domain explicitly. We propose representing the domain as an Averaged PRM using only “schemalevel” statistical information about the objects and relations, and present an approximation algorithm for reasoning about the domain with only this information. We present experimental results showing that interesting inferences can be made about extremely large domains, with a running time that does not depend on the number of objects.