TITLE:  Model Uncertainty and Robust Stochastic Modeling

ABSTRACT:

Virtually any performance analysis in stochastic modeling relies on input model assumptions that, to some extent, deviate from the truth. This talk will investigate a robust framework to quantify these model errors, by positing worst-case optimizations over the input probability distributions subject to the modeler's partial, nonparametric information. We illustrate these optimization formulations in several stylized contexts in stochastic modeling, describe their computational challenges, and present some simulation-based machinery in approximating their solutions. We also illustrate their statistical connections with conventional input modeling in the stochastic simulation literature.