Towards Dynamic Graph Neural Networks

Abstract

This thesis broadens the existing understanding of GNNs beyond a static perspective to encompass dynamic graphs, introducing novel and practical methodologies for modeling both continuous-time and discrete-time dynamic graphs. Furthermore, it proposes a theoretical framework, which serves as a crucial component in completing the theoretical underpinnings of dynamic graph neural networks. The contributions made in this thesis not only deepen the understanding of dynamic graph neural networks but also lay the groundwork for developing an extensive range of GNN-based models for real-world dynamic graphs.