Paper from School of Data Science Researchers Accepted by Prestigious Machine Learning Conference
The University of Virginia’s School of Data Science is pleased to announce that a paper from four of its researchers has been accepted by the International Conference on Machine Learning, or ICML 2024, one of the most prestigious annual artificial intelligence and machine learning gatherings.
The lead authors of the paper are Phong Nguyen, an assistant professor of data science; Shahab Azarfar, a postdoctoral research associate; and Xinlun Cheng, an incoming postdoctoral research associate. Stephen Baek, a Quantitative Foundation Associate Professor of Data Science, is the corresponding author.
They collaborated with Pradeep Seshadri, Yen Nguyen, and H.S. Udaykumar of the University of Iowa as well as Munho Kim and Sanghun Choi of South Korea’s Kyungpook National University.
The group’s work addresses the urgent issue of how to model extreme and unsteady physics challenges, such as propellant combustion and turbulent airflow, among others.
Recently, scientists have begun to use artificial intelligence and deep learning to help predict these types of phenomena. However, AI models are limited by their inability to discern whether their predicted outputs are scientifically possible.
To address this shortcoming, the research team built off their previous work on physics-aware recurrent convolutions, known as PARC, creating a version, dubbed PARCv2, that can model more intricate dynamics factors.
The newly enhanced version of PARC was able to mimic how mathematical relationships about the physics of a system, known as partial differential equations, are solved in physics simulation solvers — a breakthrough that proved critical in bridging the gap between the powerful predictive capabilities of modern AI tools and the fundamental scientific needs of physics simulations.
ICML 2024 received more than 9,000 submissions this year, accepting just 27.5% for presentation. The conference will be held July 21-27 in Vienna, Austria.
