Statitsical relational models have been successfully used to modelstatic probabilistic relationships between the entities of the domain.In this talk, we illustrate their use in a dynamic decison-theoreticsetting where the task is to assist a user by inferring his intentionalstructure and taking appropriate assistive actions. We show that thestatistical relational models can be used to succintly express thesystemu27s prior knowledge about the useru27s goal-subgoal structure andtune it with experience. As the system is better able to predict theuseru27s goals, it improves the effectiveness of its assistance. We showthrough experiments that both the hierarchical structure of the goals and the parameter sharing facilitated by relational models significantlyimprove the learning speed.
展开▼
