DDPS | Defining Foundation Models for Computational Science: Toward Clarity and Rigor скачать в хорошем качестве

DDPS | Defining Foundation Models for Computational Science: Toward Clarity and Rigor

DDPS Talk date: October 2nd, 2025 Speaker: Youngsoo Choi (LLNL, https://people.llnl.gov/choi15) Abstract: Foundation models have already reshaped fields like natural language processing and computer vision, powering tools such as ChatGPT and enabling broad generalization across diverse tasks. The excitement is now spilling into computational science, where the potential impact is enormous, spanning fluid dynamics, materials design, climate modeling, and beyond. But what exactly qualifies as a foundation model in this domain? Without a shared definition, the term risks becoming a buzzword rather than a rigorous scientific concept. In this talk, I will present our recent position paper that seeks to address this gap. I will outline a set of necessary and desirable characteristics for foundation models in computational science, emphasizing principles such as generalization, reusability, and scalability. I will also introduce the Data-Driven Finite Element Method (DD-FEM) as one concrete path toward realizing these ideas, combining the modular structure and mathematical rigor of classical numerical methods with the adaptability of modern machine learning. The goal is not to prescribe a single solution, but to spark community-wide discussion and help establish standards for evaluating and developing foundation models in high-consequence scientific domains. I will close by sharing open questions and opportunities for collaboration, as we collectively shape what foundation models should mean for computational science. Bio: Youngsoo is a staff scientist at LLNL’s CASC group, where he develops efficient foundation models for computational science. His research focuses on creating surrogates and reduced-order models to accelerate time-critical simulations in areas such as inverse problems, design optimization, and uncertainty quantification. He has pioneered advanced ROM techniques—including machine learning-based nonlinear manifolds, space-time ROMs, component-wise ROM optimization, and latent space dynamics identification—and currently leads the libROM team in data-driven surrogate modeling. His contributions extend to open-source projects such as libROM, pylibROM, LaghosROM, ScaleupROM, LaSDI, WLaSDI, tLaSDI, gLaSDI, NM-ROM, DD-NM-ROM, and GappyAE. Youngsoo earned his BS from Cornell and his PhD from Stanford, and he was a postdoc at Sandia National Laboratories and Stanford University before joining LLNL in 2017. 📚 DDPS seminar is organized by libROM team ( www.librom.net ). 🗓️ DDPS seminar schedule: https://www.librom.net/ddps.html 💻 LLNL News: https://www.llnl.gov/news 📲 Instagram: https://www.instagram.com/livermore_l... Facebook:   / livermore.lab   🐤 Twitter: https://x.com/livermore_lab/ About LLNL: Lawrence Livermore National Laboratory has a mission of strengthening the United States’ security through development and application of world-class science and technology to: 1) enhance the nation’s defense, 2) reduce the global threat from terrorism and weapons of mass destruction, and 3) respond with vision, quality, integrity and technical excellence to scientific issues of national importance. Learn more about LLNL: https://www.llnl.gov/. Release number: LLNL-VIDEO-2012305