Unique Challenges in Physics-Informed Machine Learning

Explore the unique challenges in physics-informed machine learning (PIML) through this 55-minute talk by Jie Bu at the Alan Turing Institute. Delve into the growing field of PIML and understand why it remains less understood compared to conventional machine learning areas like computer vision and natural language processing. Examine two key problems encountered in PIML: the additional non-convexity introduced by customized physics-informed loss functions, and the insufficient model expressibility when using models designed for data-driven machine learning. Gain insights into the discrepancies between well-established conclusions in conventional machine learning and the distinct aspects of PIML, highlighting the need for specialized approaches in this emerging field.