KAAI Workshop on Foundation Models & Agentic Methods for Astrophysical Simulation

Astrophysics and cosmology increasingly rely on large-scale simulations to connect theory with noisy, high-dimensional, unstructured observations. Two developments are making that connection more tractable. Foundation models, which learn transferable representations from large, heterogeneous datasets, are a natural fit for representing, compressing, and analyzing astrophysical data. Agentic methods, meanwhile, are beginning to reshape scientific workflows, expanding analysis capacity while raising open questions about how such tools can be used responsibly in astronomical research. Both directions face challenges: foundation models are limited by scarce ground-truth data and simulations, while agentic methods must contend with tasks that are often non-verifiable or computationally expensive. This workshop brings together researchers working on both fronts to discuss these challenges and where the field is headed. Foundation models for astrophysics: representing, manipulating, compressing, comparing, and interpreting astrophysical data, from simulations to observations. Agentic science loops: how agentic methods can responsibly augment the astronomical research process. Registration is free. Because in-person capacity is limited, registrations will be approved on a rolling basis. We encourage you to register early to guarantee your spot. The form will remain open through Monday, August 24. Organizing committee: Matthew Ho (Columbia University, Department of Astronomy); Barnabás Póczos (Carnegie Mellon University, Machine Learning Department); Rupert Croft (Carnegie Mellon University, Department of Physics); Tiziana Di Matteo (Carnegie Mellon University, Department of Physics); Hy Trac (Carnegie Mellon University, Department of Physics); Rachel Mandelbaum (Carnegie Mellon University, Department of Physics).