Prof. Danai Koutra receives NSF grant for research in graph neural networks

The project aims to advance the theoretical underpinnings of the interplay between graph heterophily and overall performance of graph neural networks.
Prof. Danai Koutra
Prof. Danai Koutra

Morris Wellman Faculty Development Professor Danai Koutra has received an NSF grant to fund research in graph neural networks (GNNs), which translate the success of deep learning to graph-structured data and has led to state-of-the-art results in various applications from recommendation systems and fraud detection to medicine to finance. “Collaborative Research:  III: Medium: Graph Neural Networks for Heterophilous Data: Advancing the Theory, Models, and Applications” aims to generalize GNNs to work effectively in an even wider range of domains while rectifying negative consequences to GNNs with a smaller scope. 

In the various applications of GNNs, homophily, the extent to which similar nodes of a network tend to attach to each other, presents an unknown variable that cannot be computed empirically due to limited labeled data. GNNs tailored to homophilous graphs can often produce biased, unfair, or erroneous results when applied to heterophilous data which represents certain real-world settings where dissimilar nodes attach.

The inability of GNNs to generalize their strong performance on homophilous or assortative graphs to many heterophilous graphs has attracted significant attention, and has led to empirical demonstration of the existence of “good heterophily,” where GNNs can perform well. Koutra’s project intends to focus on “robustness, fairness, and explainability [that] will help support accountable algorithmic decision-making in the domains where GNN models are employed.” 

In order to advance the theoretical underpinnings of the interplay between different types of heterophily and GNNs, Koutra hopes to provide a new theory to formally characterize the heterophily-related challenges of GNNs and a deeper understanding into “good” and “bad” heterophily, and enhance our understanding of “good” types of heterophily, which some architectures can model effectively, but have been vastly ignored until now. 

The project will also introduce new GNN designs and architectures for strong performance across different levels and types of heterophily for better algorithmic decision-making. Additionally, Koutra emphasizes the project will go “beyond the traditional tasks and heterophilous network types investigated in the literature, and will include exploration of high-impact applications along with collaborators in academia and industry.”

In 2020, Koutra was named a Morris Wellman Faculty Development Professor for her outstanding contributions to teaching and research. She has received an NSF CAREER Award and a Rising Star Award from ACM SIGKDD. She has also earned a number of research awards throughout her career, including the 2016 ACM SIGKDD Dissertation Award for her thesis on “Exploring and Making Sense of Large Graphs,” an honorable mention for the SCS Doctoral Dissertation Award (CMU), an ARO Young Investigator Award (2018), an Adobe Data Science Research Faculty Award (2018), an Amazon Research Faculty Award (2019), a WSDM 2019 Outstanding PC Award (2019), and several best paper awards and nominations.

Danai Koutra; Databases and Data Mining; Research News