Title: The Value of an Approximate Solution in Stochastic Programming with Recourse

Derek Holmes

Abstract of Recent Presentation:

Explicitly considering uncertainty in mathematical programming involves several modeling decisions. Due to the difficulty of solving stochastic programs, simplifications are frequently used. Measuring the costs of making these simplifications gives modelers quantitative tools for guiding and evaluating the modeling process. For example, the Expected Value of Perfect Information (EVPI) measures the relative benefits of eliminating uncertainty. We use a taxonomy of such measures to define the Value of an Approximate Solution. The VAS is the (potential) cost of using a suboptimal decision in terms of bounding functions on the true optimal cost. An algorithmic approach to bounding the VAS is presented. Primal and dual solution techniques will be presented, as well as refinement procedures. Click here to go back!