Training Next Generation AI Users and Developers at NCSA

TL;DR

The paper addresses building an AI-capable workforce within a high-performance computing context at NCSA by presenting FoDOMMaT, a 10-week NSF REU that pairs undergraduates with domain and ML mentors to develop open-source ML tools on HAL/Delta infrastructures. It details a structured training and project pipeline, ranging from baseline software and data practices to deep learning with PyTorch and reimplementation of landmark models, supplemented by seminars and a final symposium. An external evaluation framework provides formative and summative feedback to guide program improvements, while documented outcomes include increased diversity, publications, and continued student engagement in AI research. The findings demonstrate that a mentor-led, cohort-based REU, integrated with robust infrastructure and open-source practice, can effectively train next-generation AI researchers and contribute to the HPC AI ecosystem with tangible educational and scholarly benefits.

Abstract

This article focuses on training work carried out in artificial intelligence (AI) at the National Center for Supercomputing Applications (NCSA) at the University of Illinois Urbana-Champaign via a research experience for undergraduates (REU) program named FoDOMMaT. It also describes why we are interested in AI, and concludes by discussing what we've learned from running this program and its predecessor over six years.