Charles.Wiecha@ISL1.RI.CMU.EDU.UUCP (01/27/87)
Influence Diagrams: Graphical Representations for Uncertainty Ross D. Shachter Department of Engineering-Economic Systems Stanford University Wednesday, January 28 2:30-4:00 PM Porter Hall 223D The influence diagram is a network for structuring bayesian decision analysis problems. The nodes represent uncertain quantities, goals, and decisions, and the arcs indicate probabilistic dependence and the observability of information. The graphical heirarchy promotes discussion by emphasizing the structure of a problem and the relationships among variables, while allowing the details of assessment to be completed later. Because the components have a basic mathematical interpretation, even a qualitative diagram has a precise meaning. When the quantitative information is complete, the influence diagram can be evaluated in a generalization of decision tree solving. Examples using influence diagrams will be drawn from decision analysis, information theory, dynamic programming, Kalman filtering, and expert systems. In the latter, we ask the question "Why do probabilists insist on looking at everything backwards?"