Reduced-order modeling of high-dimensional systems using scientific machine learning

(Please email sherinetiew@nus.edu.sg to register for the seminar.)

Abstract

In this talk, I will present recent research that builds fast and accurate reduced-order models (ROMs) for various high-dimensional systems. These systems may be steady-state, where the ROM is tasked with making predictions given varying parametric inputs, or they may be dynamic where the ROM must make accurate forecasts in time, given parameters and/or varying initial and boundary conditions. In both endeavors, we will outline the development of scientific machine learning strategies, based on deep learning-based compression and forecasting, to dramatically improve accuracy and time-to-solution for extended computational campaigns. Furthermore, in addition to canonical experiments, our algorithms will be demonstrated for several real-world applications of strategic importance. Some examples are building ROMs for geophysical forecasting from ship and satellite observation data and wind-turbine wake predictions from meteorological and LIDAR measurements.

About the Speaker

Romit Maulik is an Assistant Computational Scientist in the Mathematics and Computer Division at Argonne National Laboratory.  He is also a Research Assistant Professor in the Department of Applied Mathematics at the Illinois Institute of Technology, Chicago. He was previously the Margaret Butler Postdoctoral Fellow in the Argonne Leadership Computing Facility. Before coming to Argonne, he received his PhD in Mechanical & Aerospace Engineering at Oklahoma State University. His research interests are centered around scientific machine learning algorithm development for various computational problems arising in the aerospace, geophysical, and fusion applications.

(All are welcome to attend.)