Project Grant 2619079

The National Science Foundation Division of Electrical, Communications and Cyber Systems awarded The University of Texas at San Antonio $120,000 on January 1, 2027, under the Engineering program (CFDA 47.041) to develop a subspace optimization framework for resource-efficient and privacy-preserving fine-tuning of large language models.

The research addresses computational and data-sensitivity barriers that constrain small organizations, universities, healthcare institutions, and resource-limited users from adapting general-purpose LLMs to specialized tasks. The project investigates methods to achieve performance comparable to full-parameter fine-tuning while operating within limited computing resources and protecting sensitive training data that cannot be shared. The framework dynamically constructs and updates low-dimensional subspaces that capture structured patterns in model parameter changes during fine-tuning, reducing both financial and energy costs of model adaptation.

The project runs through December 31, 2029, with performance at UTSA's San Antonio, Texas location. Work includes development of educational content, laboratory modules, and hands-on research experiences for undergraduate and graduate students. UTSA also conducts pre-college outreach introducing middle- and high-school students to modern AI and engineering through project-based learning at participating institutions.