This National Science Foundation project grant of $300,000 supports research at Michigan State University to advance graph neural network models for heterophilous data under the Computer and Information Science and Engineering program. The award period is from October 1, 2022 to September 30, 2026.
The project aims to generalize graph neural networks to work effectively on a wider range of domains characterized by heterophily, or low homophily. It will introduce new graph neural network designs and architectures that perform strongly across different levels and types of heterophily while also ensuring robustness, fairness, and transparency. Beyond traditional tasks and network types, the research will explore applications in collaboration with academic and industry partners. The funding will also support training of undergraduate and graduate students through integration in courses and research opportunities.