Abstract
Simulation models are often computationally expensive to execute. Metamodeling is a technique to approximate simulation models to support fast performance evaluation and decision making. The basic concept is that the user executes the simulation model only at a small number of carefully selected ‘‘design points’’. A metamodel can be built to approximate the true response surface by interpolating the simulation outputs. The responses at other points are then predicted by the metamodel without running the simulation at all. However, existing metamodels generally treat the simulation model as a black box, discarding the structural properties of the response surface. Therefore, they often fail to capture highly nonlinear response surfaces. In this talk, new techniques will be discussed to address this issue, including stylized models and regularization in machine learning.
Xiaowei Zhang
Associate Professor
My research research focuses on methodological advances in stochastic simulation and optimization, decision analytics, and reinforcement learning, with applications in service operations management, FinTech, and digital economy.