Logical Characterizations of the Expressiveness and Decidability of Graph Neural Networks
The authors establish logical characterizations of the expressiveness of certain classes of graph neural networks (GNNs), and use these characterizations to determine the decidability of key verification problems for GNNs.