Chun-Hung Chen received his Ph.D. degree from Harvard University in 1994. He is currently a Professor at George Mason University. Dr. Chen was an Assistant Professor at the University of Pennsylvania before joining GMU. He was also affiliated with National Taiwan University (Electrical Eng. and Industrial Eng.) from 2008-14. Sponsored by NSF, NIH, DOE, NASA, MDA, Air Force, and FAA in US, he has worked on the development of very efficient methodology for stochastic simulation optimization and its applications to air transportation system, semiconductor manufacturing, healthcare, security network, power grids, and missile defense system. Dr. Chen received a “National Thousand Talents Award” (国家千人) in 2011, the Best Automation Paper Award from the 2003 IEEE International Conference on Robotics and Automation, 1994 Eliahu I. Jury Award from Harvard University, and the 1992 MasPar Parallel Computer Challenge Award. He has served as a Department Editor for IIE Transactions, Department Editor for Asia-Pacific Journal of Operational Research, Associate Editor for IEEE Transactions on Automation Science and Engineering, Associate Editor for IEEE Transactions on Automatic Control, Area Editor for Journal of Simulation Modeling Practice and Theory, Advisory Editor for International Journal of Simulation and Process Modeling, and Advisory Editor for Journal of Traffic and Transportation Engineering. Dr. Chen is the author of two books, including a best seller: “Stochastic Simulation Optimization: An Optimal Computing Budget Allocation”.

 

Title of Speech:  Multi-fidelity Optimization for Engineering Design with Ordinal Transformation and Optimal Sampling

 

Abstract:  Simulation/evaluation models at different fidelity levels are often available for analyzing and designing complex systems. High-fidelity models can accurately predict the performance of a design alterative but are time-consuming to run. Low-fidelity models are much faster but usually lead to bias and variability. Multi-fidelity optimization provides a means to achieve design optimization at reduced computational cost by using a high-fidelity model in combination with lower-fidelity models. However, most existing methods are limited to problems with relatively smooth response with low-dimensional design space. We propose an innovative way of utilizing the lower-fidelity model, called “ordinal transformation”. The new method can reduce a multi-dimensional and even categorical design space into one-dimensional space which is smoother and has nice properties. Utilizing the idea of Optimal Computing Budget Allocation (OCBA) invented by the speaker, we further develop a novel sampling scheme that can optimally determine which designs in the new space to evaluate using the expensive but accurate high-fidelity model. The goal is to maximize the overall optimization efficiency. At the end, we show the effectiveness of our approach by applying it to a machine allocation problem in semiconductor manufacturing.