Graph Neural Networks (GNNs) have excelled in diverse applications due to their outstanding performance, demonstrating remarkable capabilities in tasks ranging from node classification to graph generation. However, despite their success, GNNs could inherit and exacerbate existing societal biases in their decision-making processes, posing a significant concern given their widespread deployment in critical systems. Consequently, there has been a surge in efforts to tackle fairness issues and ensure GNNs promote equitable outcomes. However, most of them rely on the statistical fairness notions, which assume that biases arise solely from sensitive attributes, neglecting the pervasive labeling bias prevalent in real-world scenarios. In addition, existing approaches usually focus on a single notion of fairness, ignoring the interactions between different fairness goals. Moreover, fair classification tasks have been the primary focus of research, while biases in the increasingly prevalent graph generative models remain largely unattended. To this end, this project aims to develop a new and versatile fair graph learning framework that can: 1) accurately identify the various biases present in the graph, 2) simultaneously address multiple biases in both graph classification and graph generation tasks, 3) improve fairness by decomposing sensitive information in node representations, while retaining task-related information, 4) generate diverse underrepresented samples and establish fair link connections to ensure consistent representation across various groups. The research approaches include three primary research objectives: i) identify real counterfactual instances directly from the dataset to guide the bias mitigating process, ii) achieve individual and group fairness simultaneously, and iii) the first of its kind fair graph generation methodology.

Zichong Wang is a third-year Ph.D. student at the Knight Foundation School of Computing and Information Sciences, Florida International University, under the supervision of Dr. Wenbin Zhang. His research focuses on mitigating inadvertent disparities caused by the interaction of algorithms, data, and human decisions in policy development. His work has garnered significant recognition, including the Best Paper Award at FAccT 2023 and a nomination for the Best Paper Award at ICDM 2023, with over 10 top-tier publications in leading venues. He also serves as the Web Chair for WSDM 2024 and actively contributes as a Program Committee member and reviewer for conferences and journals like AAAI, IJCAI, and Machine Learning. More information can be found at: https://lavinwong.github.io/