Learning Influence Diagram Utility Function by Observing Behavior

Abstract

This paper considers the task of learning utility functions of certain influence diagram based on the decision maker’s past decisions. We assume that the influence diagram structure and the probability distribution it assigns to random events are known, so that we need only infer the utility function u for its. We also assume that the decision maker is rational. In particular, the decision maker’s past decisions can be viewed as constraints on u. So, if we have a prior probability distribution p(u) over u, we can then condition on these constraints to obtain u. In this paper, an approach for learning utility functions from decision maker’s behavior was proposed. We also show that it is effective.