Project Grant 2551752

Federal Grant Award Summary

The University of North Carolina at Chapel Hill received a $300,000 Project Grant from the National Science Foundation's Division of Computing and Communication Foundations under the Computer and Information Science and Engineering (CFDA 47.070) program, effective September 1, 2026 through August 31, 2029. This collaborative research initiative develops a knowledge augmentation framework to improve model editing in foundation models, such as large language models (LLMs), enabling more reliable, generalizable, and trustworthy updates to outdated or incorrect knowledge. The project addresses critical gaps in existing model editing methods by researching knowledge augmentation approaches during both model training and inference, including visual generation and text augmentation for training-time editing, and contrastive decoding and preference-based optimization for inference-time editing.

The research will evaluate proposed methods across diverse applications including text-to-image generation, question answering, classification, and completion tasks, with pilot applications in medical image diagnosis and geographic remote sensing. In addition to advancing the scientific understanding of how foundation models store, update, and apply knowledge, the project incorporates substantial education and outreach activities designed to train undergraduate and graduate students in reliable foundation models, multimodal learning, and responsible machine learning, with deliberate efforts to broaden participation in computing. Successful completion will support safer and more trustworthy deployment of foundation models across critical domains such as healthcare and science.