ABSTRACT

This talk will give an overview of plausible inference, a nascent mathematical framework that leverages limited simulation experiments and known or assumed functional properties of performance measures to deliver statistical inferences on the performance of unsimulated systems. Methods of plausible inference essentially formulate and solve optimization problems at systems of interest to determine plausible values for their performances. The results can be used to produce confidence intervals or screen out systems with unacceptable performance for the purposes of optimization, feasibility determination, or comparison to a target. These inferences come with guarantees of uniform confidence and consistency and their power depends on the functional properties being exploited, such as monotonicity, Lipschitz continuity, and convexity. The framework can also be extended to incorporate stochastic gradients and handle multiple performance measures, as arises in simulation-optimization problems with stochastic constraints or multiple objectives.

BIOGRAPHY

David J. Eckman is an Assistant Professor in the Wm Michael Barnes '64 Department of Industrial and Systems Engineering at Texas A&M University. He received a Ph.D. in Operations Research in 2019 from Cornell University and was a postdoctoral scholar at Northwestern University from 2019-2021. His research interests deal with optimization and output analysis for stochastic simulation models. He is a co-creator of SimOpt, a testbed of simulation-optimization problems and solvers, and is a council member of the INFORMS Simulation Society.