Scientific Machine Learning - From Physics-Informed to Data-Driven

Explore the rapidly expanding field of scientific machine learning (SciML) in this comprehensive seminar that bridges scientific computation and machine learning methodologies. Learn about the evolution from physics-informed approaches like Physics-Informed Neural Networks (PINN) and Physics-Informed Gaussian Processes (PIGP) to modern data-driven operator learning paradigms. Discover how collocation methods enforce differential equations at grid points and understand the emerging "physics is in the data" philosophy of operator learning for parametric differential equation models. Examine the advantages and disadvantages of both physics-informed and data-driven approaches through practical examples, including recent applications to the Hasegawa-Wakatani model of plasma turbulence in tokamak fusion devices. Gain insights from Professor Ulisses M. Braga-Neto of Texas A&M University, founding Director of the Scientific Machine Learning Lab at TAMIDS, as he shares cutting-edge research developments and real-world applications in this dynamic field that combines theoretical physics with advanced machine learning techniques.