Time-Aware Robustness of Temporal Graph Neural Networks for Link Prediction (Extended Abstract)

Marco Sälzer, Silvia Beddar-Wiesing

Abstract: We present a first notion of a time-aware robustness property for Temporal Graph Neural Networks (TGNN), a recently popular framework for computing functions over continuous- or discrete-time graphs, motivated by recent work on time-aware attacks on TGNN used for link prediction tasks. Furthermore, we discuss promising verification approaches for the presented or similar safety properties and possible next steps in this direction of research.