SDM 25' Tutorial

Many real-world networks evolve over time, which results in dynamic graphs such as human mobility networks and brain networks. Usually, the “dynamics on graphs” (e.g., node attribute values evolving) are observable, and may be related to and indicative of the underlying “dynamics of graphs” (e.g., evolving of the graph topology). Tradition- ally, both types of dynamics have been highly complex to model, but in recent years, research on each has attracted increasing attention with the help of expressive models, such as graph neural networks. However, modeling and mapping between them have not yet been well-explored on account of two significant challenges: 1) the difficulty with mapping continuous multi-attributed data to discrete graph topologi- cal data, and 2) the ineffectiveness and inefficiency of learn- ing the long, high-dimensional time series of the dynamics. To address these challenges, we propose to learn the map- ping between the dynamics on graphs and the dynamics of graphs with a new graph neural network, which is based on an adaptive echo-state architecture. Specifically, we pro- pose a new deep echo-state graph dynamics encoder that is scalable for graphs with large time duration and sizes. Then to ensure a good learnability, a new neural architec- ture search (NAS) strategy is proposed and tailored for the proposed deep echo-state encoder and dynamic graph pre- dictor. Extensive experiments on synthetic and real-world application data demonstrate the outstanding effectiveness and efficiency of the proposed method.

May 3, 2025 1:30 PM — 3:30 PM