Despite achieving remarkable performance, deep graph learning models, such as node classification and network embedding, suffer from harassment caused by small adversarial perturbations.

Inemergent communication studies, one often considers agents who canshare information about a set of objects described by the common features.