This $600,000 Project Grant awarded by the National Science Foundation's (NSF) Computer and Information Science and Engineering (CISE) program (CFDA 47.070) aims to enhance machine learning on graph-structured data to address the challenge of data distribution shifts in real-world artificial intelligence applications.
The research project focuses on two main thrusts: 1) developing methods to estimate and mitigate shifts in graph structure from training to evaluation, targeting tasks like node classification and link prediction, and 2) creating foundational graph data models that are provably expressive, generalizable, and scalable, while investigating robust fine-tuning techniques. The methodologies will be evaluated on applications in particle physics and molecular representation. In addition to the technical research, the project emphasizes knowledge dissemination through new datasets, software tools, workshops, and student engagement, particularly targeting underrepresented groups.
Generated 2/25/25, 5:27 AM