Dynamic Graph Learning

Dynamic graph learning studies networks whose connections, attributes, or both evolve, from timestamped interaction events to sequences of graph snapshots. Common approaches include temporal graph neural networks with memory, attention, or state-space models, alongside methods that learn changing graph structure, temporal representations, or causal patterns. They are used for tasks such as link and event prediction, anomaly detection, traffic forecasting, and analysis of brain or transaction networks. A central challenge is balancing responsiveness to new patterns with retention of useful history, while ensuring efficient, interpretable models are evaluated under realistic distribution shifts and negative-sampling protocols.