Subgroup Generalization and Fairness of Graph Neural Networks

Based on this analysis, our second contribution is the discovering of a type of unfairness that arises from theoretically predictable accuracy disparity across some subgroups of test nodes.