Understanding the Expressivity of Causality-Aware Graph Neural Networks for Temporal Graphs

Vortrag von Prof. Dr. Ingo Scholtes

Datum: 26.03.26  Zeit: 12.00 - 13.00  Raum: Y27H12

Graph Neural Networks (GNNs) have become a cornerstone for the application of deep learning to relational data on complex networks. However, we increasingly have access to time-resolved data on temporal graphs that not only capture which nodes are connected to each other, but also when and in which temporal order those connections occur. A number of works have generalized GNNs to temporal graphs, but our understanding for the expressivity of these models remains limited.

Addressing this gap, we propose a novel notion of temporal graph isomorphism and develop a temporal generalization of the Weisfeiler-Leman algorithm to heuristically distinguish non-isomorphic temporal graphs. Building on this foundation, we derive a novel message passing scheme for temporal graph neural networks that operates on the event graph representation of temporal graphs. Our approach accounts for temporal-topological patterns that unfold via causal walks, i.e. temporally ordered sequences of connections by which nodes can causally influence each other over time.