Foundation Models for Computational Science

The focus of this project will be on the inner loop of computational sciences, which includes hypothesis creation and evaluation, carried out using various high-fidelity simulation codes and workflows. This process iteratively generates detailed and accurate computational models to evaluate hypotheses, ensuring precise and reliable results through repeated, fine-grained calculations and data processing. This project aims to develop a sophisticated personal assistant for computational science. In this framework, the scientist interacts with a computational science foundation model (CompSciFM), which is capable of producing code and workflow integrating with large-scale highly-performant codes, which gets executed through the HPC agent.