Flexible Policy Construction by Information Reenement

We report on work towards exible algorithms for solving decision problems represented as innuence diagrams. An algorithm is given to construct a tree structure for each decision node in an innuence diagram. Each tree represents a decision function and is constructed incrementally. The decision maker can balance the cost of computing the next incremental improvement to a tree against the expected value of the improvement. The improvements to the tree converge to the optimal decision function computed by dynamic programming techniques, and the asymptotic behaviour is only a constant factor worse than dynamic programming techniques, counting the number of Bayesian network queries. Empirical results show that utility varies with the size of the tree and the number of Bayesian net calculations.