Knowledge Representation in Decision Analysis

SC12.1 Stochastic Tree Modeling Gordon B. Hazen, Jay Sounderpandian, James Pellissier --- Northwestern Univ., IE/MS Dept., Evanston, IL 60208-3119,

Stochastic trees are graphical tools which marry decision trees and¨ Markov chain transition diagrams. We have used them to model medical¨ treatment decisions. We discuss stochastic tree basics and present¨ STOTREE, a graphical modeling spreadsheet add-on for the formulation¨ and solution of multiple-factor stochastic trees.

SC12.2 Application of Chain-Graph Models to Decision Analysis Pierre Nidilikilikesha --- Duke Univ., Fuqua Sch. of Bus., Durham, NC 27708-0120, (ndili@mail.duke.edu)

I study chain-graph models involving decision variables. Unlike¨ influence diagrams, chain-graph models contain both directed and¨ undirected arcs. There is a practical need for chain-graph¨ representations in decision analysis because the dependence¨ relations among variable in many decision problems to not fit easily¨ into the influence-diagram representation...

SC12.3 A Forward Monte Carlo Method for Solving Influence Diagram Using Local Computation John M. Charnes, Prakash P. Shenoy --- Univ. of KS, Sch. of Bus., Summerfield Hall, Lawrence, KS 66045-2003, (j-charnes@ukans.edu)

We investigate a forward Monte Carlo sampling technique that draws¨ independent and identically distributed observations. Methods that¨ have been proposed in this spirit are sampled from the entire¨ distribution. However, when the number of variables is large, the¨ state space of all variables is exponentially large and the sample¨ size required for good estimates is too large to be practical...