Learning models of macrobehavior in complex adaptive systems

Computational systems often exhibit complex aggregate behaviors, called macrobehaviors, which arise over time from interactions of the individual entities in the system. Macrobehaviors are often not well understood and are difficult to identify and predict without automated tools; however, to date there are very few statistical tools designed for this purpose. In this thesis, I will develop a new class of models for relational data called compositional models that consider the characteristics of aggregations of individuals in addition to the individual attributes and structure present in the data.