15h30 - 17h10
Surrogate Modeling for High-Dimensional Simulation-Based Optimization Problems
Motivated by challenges that arise in tackling high-dimensional, continuous and discrete, stochastic simulation-based optimization problems for urban mobility, we review past work on the use of analytical surrogate modeling. We discuss the use of surrogates to design exploitation techniques, exploration techniques, and combined exploitation-exploration techniques. Discussed topics include metamodeling, sampling in high-dimensional spaces, and Bayesian optimization. We present their application for both model calibration and traffic management problems with case studies of major metropolitan areas, including Singapore and New York City.