ABSTRACT
We will begin with utilities and then describe how they are represented together with probabilities in decision networks. Then we present the algorithm for evaluating a decision network to make individual decisions, illustrating with several examples. We are also interested in determining the best sequences of decisions or actions, that is to say, with planning. First, we will use a decision network for a “test-thenact” combination of decisions. Then, we introduce dynamic Bayesian networks for explicitly modeling how the world changes over time. This allows us to generalize decision networks to dynamic decision networks, which explicitly model sequential decision making or planning under uncertainty. We conclude the chapter with a description of object-oriented Bayesian Decision Networks, a framework for building large, complex, hierarchical Bayesian decision networks.
