Addressing Input Model Uncertainty in Stochastic Simulation: From Quantification to Optimization

Georgia Tech Research Corporation, Atlanta GA

Investigators

Abstract

Simulation and optimization techniques are often used to evaluate system performance and facilitate decision making in complex and stochastic systems. This project aims to quantify the risk associated with simulation modeling and analysis, and to design robust and risk-aware strategies for decision making based on simulation. Because of the generality of the proposed approaches, the resulting techniques will have broad applicability in a wide array of industry and science sectors. Through collaborations with researchers in industry, the developed algorithms will be tested on and applied to problems in the application area of sharing economy. This project will also provide training opportunities for underrepresented groups through recruiting and outreach activities. Stochastic simulation is often used for performance analysis and decision making in complex systems. The input to the simulations is a collection of distributions based on data, and uncertainty in the input brings significant risk to decision making. The goal of this project is to develop theory and methods that quantify the risk associated with input uncertainty, support decision making systems that are robust to the risk associated with the input uncertainty, and handle streaming data which arrive sequentially in time. The project consists of three major research thrusts including a) online quantification of input model uncertainty developed with convergence guarantees for both parametric and non-parametric input models, b) simulation optimization under input model uncertainty within a new framework of Bayesian risk optimization which aims to balance optimizing the expected performance and hedging against the input model risk, and c) ranking and selection under input model uncertainty for which new algorithms will be developed with convergence rate results; these will take into account the trade-off between reducing the input uncertainty via collecting more data and reducing the simulation uncertainty by running more simulation experiments. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.