AFOSR: Simulation-Based and Risk-Sensitive Methodologies for Stochastic Optimization and Control

This research will study basic questions aimed at challenges in information superiority, logistics, and planning. The researchers will develop and analyze new algorithms for the simulation optimization approach of sequential response surface methodology by incorporating direct gradient estimates; develop and analyze new global stochastic kriging simulation metamodels using an extrapolation method enabled by direct gradient estimates; utilize risk-sensitive cost functions to achieve express risk preferences and robustness in control problems; study how incorporation of risk-sensitivity affects the behavior of decision makers and controllers; develop and study efficient sampling and simulation-based methods for risk-sensitive control problems; study population-based methods for finding and improving on a good set of policies in risk-sensitive problems; and apply these optimization methodologies to practical problems, such as preventive maintenance, path planning for unmanned aerial vehicles, data mining, supply chain management, and financial engineering

This is a three-year, $554K grant.