Reference: Breese, J. S. & Horvitz, E. J. Ideal Reformulation of Belief Networks. Association for Uncertainty in Artificial Intelligence, Cambridge, MA, 1990.
Abstract: The intelligent reformulation or restructuring of a belief network can greatly increase the efficiency of inference. However, time expended for reformulation is not available for the primary task of performing inference with the network. Thus, given a cost of delay, there is a tradeoff between time dedicated to reformulating the network and time applied to the solution of the formulation. We explore the ideal partition of resources into time for reformulation and time for inference. After describing principles for computing the partition of resources under uncertainty, we discuss empirical work on belief networks. In particular, we determine the ideal amount of time to devote to searching for optimal clusters in belief networks. Given a preference model, describing the value of a solution as a function of the delay needed for its computation, our system selects an ideal time to devote to reformulation.