Abstract – Recent advances in generative AI are transforming how we model and understand complex geophysical systems. Among these, Physics-Informed Neural Networks (PINNs) offer a powerful framework that integrates governing physical laws directly into machine learning models. By embedding partial differential equations—such as those describing fluid flow, heat transport, and atmospheric dynamics—into the training process, PINNs can produce physically consistent solutions even from sparse, noisy, or incomplete observational data. This seminar will start with a basic Neural Network Application 101, then introduce the fundamentals of generative AI for scientific applications, with a focus on PINNs and their relevance to Marine, Earth, and Atmospheric Sciences (MEAS). Drawing on examples inspired by the Generative AI for Science framework and its accompanying code, we will demonstrate how PINNs can be applied to problems such as heat transport, data gap filling, climate dynamics, and subsurface transport. Compared to traditional numerical solvers, PINNs provide a mesh-free and flexible approach that is particularly well-suited for data assimilation and inverse modeling in complex, multiscale environments.

Feel free to bring your laptop if you’d like to participate in the hands-on demo.