Chih-Li Sung receives NSF CAREER Award

Chih-Li Sung, has been awarded a prestigious 2024 Faculty Early Career Development, or CAREER, Award from the National Science Foundation for his ongoing work examining the effectiveness of single-fidelity versus multi-fidelity simulations. Over the next five years, he will receive more than $420,000 to advance his research, which has the potential to transform fields such as engineering, medical science and biology. 

Computer simulations, which have tremendous predictive qualities, are essential for understanding complex systems and are used in nearly every scientific discipline. 

“Computer models are sometimes referred to as ‘digital twins’ because they serve as virtual replicas of physical systems,” said Sung, an assistant professor in the Department of Statistics and Probability at Michigan State University. “Real-time data and computer models can be used to mirror their real-world counterparts, allowing for continuous monitoring and optimal decision making.” 

Computer simulations are intended to complement and inform experiments, which are often costly and time consuming. The problem with most simulations, however, is that they use a single, high-fidelity model that is incredibly accurate, but computationally intensive. 

In other words, a highly accurate computer model based on a single, high-fidelity data source that gives excellent results requires a long time to complete a single run. In addition, the more time spent on computation set-up, the more expensive the simulation. 

“The downside of this single, high-fidelity model is that it's often very time consuming and also very difficult to set up, and sometimes it's not scalable,” said Sung. “If it’s taking you seven days to get one result, you’re not going to be getting a lot of useful insights. You don't get a chance to try different settings and to see how this model really works.” 

On the other hand, multi-fidelity models can be an efficient alternative to single fidelity models. The idea behind multi-fidelity simulations is that they combine high- and low-fidelity models to balance accuracy and computational efficiency.  

While high-fidelity simulations are more precise, they are also more expensive. Low-fidelity simulations, though less accurate, are faster. By strategically integrating these simulations, accuracy can be potentially improved without excessive computational resources. The key question is whether it is more effective to use single-fidelity or multi-fidelity simulations 

Sung and his team will work to better understand the situations where multi-fidelity modeling fails and, hopefully, what can be done to improve it in those cases. His work will investigate whether using multi-fidelity simulations can offer meaningful results that are more efficient and less costly. 

“Professor Sung has developed highly novel approaches to emulation of large complex computer experiments. These computer experiments are expensive or impractical to run to get many outputs. Professor Sung’s methods of emulation allow for effective calibration of computer experiment and uncertainty quantification, and have broad applications, from disease modeling to climate modeling to manufacturing and other areas,” said Hao Zhang, chairperson of the Department of Statistics and Probability. “This award is a recognition of the excellence of his research. I am very proud of him for his achievements.” 

“This is very challenging work, but it can be done through careful design,” Sung said. “What I mean by design is careful data selection. So, for example, I can just run these simulations sequentially. If I start with multi-fidelity simulations and discover at some point that the low-fidelity data is really inaccurate, then I would just discard the low-fidelity data and then just keep collecting and adding high-fidelity data.” 

At MSU, Sung is also interested in broadening the impact of his work by collaborating with new colleagues, such as researchers in the Department of Plant Biology. He is also working with the Department of Mathematics on an undergraduate research experience program involving exchange students. He said his booth had a good turnout at the MSU Science Festival in April. Sung has four graduate students on his team and said he is committed to recruiting more, especially students from under-represented groups. 

Sung is confident that his team’s research will enable researchers to design simulations that are more accurate and hopefully less expensive. 

“Most people tend to think the more data the better,” he said. “But this is not always true because including low-fidelity data can sometimes make the results worse. This is especially important for practitioners in engineering, healthcare and biology where they use multi-fidelity simulations all the time. Our research can really help them better understand multi-fidelity systems and how to optimally design their simulations.” 

 

• Tags:statistics and probability