Banse Early Career Scientist Seminar

Wednesday, February 14, 03:30 PM to 04:30 PM PST

Ocean Sciences Building (OSB), Room 425 with tea and cookies to follow

Andrew Shao, Hewlett Packard

Title: Two Worlds Collide: AI/ML and Scientific Simulations

Abstract: The use of AI/ML has exponentially increased in the last decade both in public-facing applications but also within scientific research. An emerging area of research focuses on how to combine AI techniques with traditional scientific simulations. Two methodologies that show specific promise are AI-in-the-loop, where the AI method is embedded within the numerical solver itself, and AI-around-the-loop, where the AI method is applied to the simulation. This talk presents a few implementations of these paradigms using HPE’s SmartSim library, an open-source software package used to build hybrid simulation/AI workflows. Two examples specifically related to the ocean will demonstrate embedded neural networks inside of MOM6 that help estimate the Gent-McWilliams coefficient and also provide a form of backscatter to mimic the inverse turbulent kinetic energy cascade. Additionally, additional demonstrations using computational fluid dynamics will be shown that demonstrate how AI can be used to find optimal tunings of simulations and be used to find control schemes to control turbulent flow. Throughout this talk, I will also be discussing how my past experience in academia and as a government scientist have helped enable successful collaborations between public science and industry.