Towards Trustworthy Graph Neural Networks: Robustness, Privacy, and Fairness

The thesis discusses the trustworthiness of Graph Neural Networks (GNNs), which are used in diverse real-world applications. Despite their performance, GNNs can have adverse effects like susceptibility to adversarial attacks, privacy risks, and potential discrimination. It proposes a comprehensive survey to answer what trustworthy GNNs are. To build trustworthy GNNs, it explores the robustness, privacy and fairness of GNNs. This thesis delves into the goals of developing an algorithm to understand GNNs' susceptibility to adversarial evasion attacks, identifying factors leading to disparate privacy vulnerability of edges, and investigating the interaction between privacy and fairness in GNNs.