ABSTRACT

Stochastic simulation models that capture the dynamics of complex systems often require a significant amount of running time. They are typically not suitable for real-time decision makings. In this paper we propose a quantile-regression based generative-metamodeling approach to learn from the simulation data and to create a fast “simulator of simulator”, which can generate observations that have the (approximately) same distribution as the original simulator, but with a much faster speed that supports real-time decision makings. Numerical experiments show that the approach works well, compared the state-of-the-art generative models. We also extend this approach to other applications, including simulating multi-dimensional economic data and generating simple images.

BIO

Jeff Hong received his bachelor's and Ph.D. degrees from Tsinghua and Northwestern University, respectively. He is currently with Fudan University, holding the positions of Fudan Distinguished Professor, Hongyi Chair Professor, Chair of Department of Management Science in School of Management, and Associate Dean of School of Data Science. He was Chair Professor of Management Sciences at City University of Hong Kong, and Professor of Industrial Engineering and Logistics Management at the Hong Kong University of Science and Technology. Prof. Hong’s research interests include stochastic simulation, stochastic optimization, machine learning, risk management and supply chain management. He is currently the Simulation Area Editor of Operations Research, an Associate Editor of Management Science and ACM Transactions on Modeling and Computer Simulation, and he was the President of INFORMS Simulation Society from 2020 to 2022