Abstract: Physics Informed Machine Learning (PIML) has emerged at the forefront of research in scientific machine learning with the key motivation of systematically coupling machine learning methods (ML) with prior domain knowledge often available in the form of physics supervision. PIML has shown great promise across a plethora of scientific applications such as quantum mechanics, turbulence modeling, climate science, and solving partial differential equations (PDEs). In this talk, I will dive deeper into two key frontier problems in PIML: (1) designing compatible PIML models for uncertainty quantification, and (2) investigating the failure mode in PIML approaches for solving PDEs. I will also briefly share several other recent advances of PIML in various application domains such as lake modeling, estimating cellular forces from phase-contrast microscopy, and methane leak detection. I will then discuss ongoing and future efforts of advancing the current state-of-the-art in PIML, highlighting the potential impact of PIML approaches on solving complex scientific problems.
Speaker’s Bio: My research is aimed at developing next-generation artificial intelligence solutions for scientific applications with the ultimate goal of accelerating scientific discovery. Specifically, I focus on incorporating scientific laws into deep learning models to improve their overall generalizability and interpretability. My research impacts socially relevant application domains such as climate science and sustainability, biological sciences, and dynamical systems governed by partial differential equations. I am also interested in solving inverse problems in scientific ML.
Last Updated: February 24, 2023 - 9:53 am