When: Wednesday, February 26, 2025 from 2pm – 3pm
Where: Amy Gutmann Hall, Room 414
Title: “Closing the Gap Between Scientific Foundation Models and Real-World Applications”
Abstract: In the era of LLM models, one gets notoriously confronted with the question of where we stand with the applicability of large-scale deep learning models within scientific or engineering domains. The talk is motivated by recent triumphs in weather and climate modeling, and discusses potentials, breakthroughs, and remaining challenges in fluid dynamics and related engineering fields. Concretely, we showcase recent work in scaling neural networks to model multi-physics phenomena and computational fluid dynamics as used in automotive engineering. Finally, we outline challenges and potential solutions when it comes to scalability beyond traditional numerical schemes and discuss the respective impact on industry and scientific environments.
Bio: Johannes Brandstetter is leading a group on “AI for data-driven simulations” at the Institute for Machine Learning at the Johannes Kepler University (JKU) Linz. Additionally, he is a Chief Researcher at NXAI – their new European AI hub in Linz (Austria). He obtained his PhD working on Higgs boson physics at the CMS experiment at CERN, and since then has worked in Deep Learning, AI4Science, and neighboring interdisciplinary fields. His interests comprise data-driven simulations, Geometric Deep Learning, and possible extensions to engineering and scientific applications.