graph neural networks (GNNs) have recently been used for node and graph
classification tasks with great success, but GNNs model dependencies among the
attributes of nearby neighboring nodes rather than dependencies among observed
node labels. In this work, we consider the task of induc