Güzin Bayraksan, an associate professor of integrated systems engineering at The Ohio State University, gave a plenary talk at the XV International Conference on Stochastic Programming (ICSP) in Trondheim, Norway, on August 2, 2019.
Bayraksan, who is a principal investigator in the Argonne-led MACSER (Multifaceted Mathematics for Rare, High-Impact Events in Complex Energy and Environment Systems) project, spoke on “Effective Scenarios in Distributionally Robust and Risk-Averse Stochastic Programs.”
A key task in developing stochastic optimization models is finding an underlying probability distribution to represent the uncertainty.
“Traditional stochastic programs assume that the probability distribution is known. In practice, however, this is often not the case,” Bayraksan said. She explained that although partial information about the uncertainty may be available from historical data, that information may have changed. In many other cases, the probability distribution is unknown and cannot be accurately approximated.
To address these situations, Bayraksan and her colleagues investigated problems where the distributional ambiguity is modeled by the total variation distance, a statistical distance metric standard in probability theory. In her talk, she reviewed conditions that can be used to measure the “effectiveness” of scenarios, and she presented a framework she and her team have developed to identify scenarios that are critical to a decision-making problem under risk and uncertainty.
“Such critical scenarios can be used in multiple ways, including providing managerial insights into important problems with complex uncertainties, enabling scenario reduction for efficient solution of large-scale problems, and guiding the selection of a risk level acceptable to decision makers,” Bayraksan said.
ICSP is the premier event of the Stochastic Programming Society and is a technical session of the Mathematical Optimization Society. The event is held every three years and brings together researchers who work on decisions under uncertainty. For more information about this year’s conference, see the website.