Dr. Ahti Salo, professor, Aalto University School of Science, will present his talk "Decision Programming for Optimizing Multi-Stage Decision Problems Under Uncertainty" on Wednesday, April 10 from 11 a.m. to noon in the CCSE Lab, Babbio 541A.
Multi-stage problems under uncertainty can be represented as influence diagrams consisting of decision, chance and value nodes connected by arcs. At each decision node, the solution can be derived either by transforming the diagram or by solving the equivalent decision tree with dynamic programming. Both approaches assume that earlier decisions are known when making later ones, which may not be the case in distributed problems. Moreover, dynamic programming is restrictive in that optimal strategy within a given branch cannot depend on the decisions in other, non-overlapping branches. Thus, the objective function cannot include risk measures such as semi-absolute deviation which would capture the variability of consequences over all branches. Interdependencies between branches arise in project portfolio selection, too, because the use of shared resources implies that decisions for one project depend on those for others.
In this paper, we develop the decision programming approach for multi-stage decision problems which can represented as extended influence diagrams without the ‘no forgetting’ assumption in the presence of multiple objectives and constraint types. This approach also extends contingent portfolio programming (Gustafsson and Salo, Oper. Res., 2005) to the selection of projects which can impact scenario probabilities. It is efficient for problems of realistic size, because the solutions can be obtained with mixed-integer linear programming (MILP).
Dr. Ahti Salo has worked extensively on the development of decision analytic methods and their uses in resource allocation, innovation, risk management, technology foresight, and efficiency analysis. He has published widely in leading international journals (including Management Science and Operations Research) and received several awards for his research from the Decision Analysis Society of the Institute for Operations Research and the Management Sciences (INFORMS).