How can new advances in machine learning impact real-world problems and discoveries?
Assistant Professor Xiaowei Jia’s research seeks to answer this question. In June 2023, Jia received the National Science Foundation (NSF) CAREER Award for his work on the project titled “Combining Machine Learning and Physics-based Modeling Approaches for Accelerating Scientific Discovery”.
The NSF CAREER Award is part of the NSF’s Faculty Early Career Development program, which supports faculty early in their career in advancing the research goals and mission of their department or school.
“This proposal is about the idea of knowledge-guided machine learning,” said Jia. “Machine learning (ML) has had success in many commercial applications, using large complex models. But there’s a growing interest in using ML to help us solve many scientific problems, like ensuring the security of water and food supplies or simulating fluid dynamics."
Jia noted that previously, industries and research groups utilized complex physical and mathematical models in their efforts to simulate various processes. However, through advances in ML, there is now hope to use machine learning models to efficiently address and study complex problems.
When asked about the broader interest in ML, Jia emphasized that his proposal requires researchers of different backgrounds to learn from each other as well as the research itself. “Even though there are a lot of people working in machine learning or AI, not too many people know or have experience in deep collaboration with domain scientists in various disciplines and creating products that benefit both AI and the domain science,” Jia said.
This proposal spans departments within the School of Computing and Information (SCI) as well as the University of Pittsburgh.
“So far, I’m working with faculty members in other departments," Jia said. This includes Peyman Givi (distinguised professor, Swanson School of Engineering), Hassan Karimi (professor, SCI), and Xu Liang (professor, Swanson School of Engineering). “One of the reasons the NSF really liked this proposal was its potential for collaborations, and how it showed promise for deploying algorithms to real systems and real problems.”
According to Jia, there are three main takeaways from this proposal.
First, this research can make an impact on other domains; ML for data-driven solutions is important for numerous applications and areas.
Second, this research will bring new innovations to enhance current ML algorithms so accuracy, robustness, and interpretation of results can be increased and models can be generalizable to different scenarios.
The third takeaway, and perhaps the most important, is trust. “If we really want these models to be useful, we have to have domain scientists trust this model. This model needs to incorporate known physical knowledge, create physically consistent results, and provide new insights to advance scientific discovery. We have to combine existing data, models, and knowledge, and make the various formats work together,” Jia said.
In addition to Assistant Professor Jia, other recipients of the NSF CAREER Award at SCI include:
- Peter Brusilovsky (Director, Intelligent Systems Program), 2005
- Panos Chrysanthis (Professor, Department of Computer Science), 1995
- Rebecca Hwa (Professor, Department of Computer Science), 2008
- James Joshi (Professor, Department of Informatics and Networked Systems), 2006
- Adriana Kovashka (Associate Professor, Department of Computer Science), 2021
- Alexandros Labrinidis (Professor, Department of Computer Science), 2008
- Adam J. Lee (Associate Dean for Academic Programs; Professor, Department of Computer Science), 2013
- Lingfei Wu (Assistant Professor, Department of Informatics and Networked Systems), 2023
- Youtao Zhang (Professor, Department of Computer Science), 2005
This news was originally shared on sci.pitt.edu here.