Defending Graph Convolutional Networks against Dynamic Graph Perturbations via Bayesian Self-Supervision

In recent years, plentiful evidence illustrates that Graph Convolutional Networks (GCNs) achieve extraordinary accomplishments on the node classification task. However, GCNs may be vulnerable to adversarial attacks label-scarce dynamic graphs. Many existing works aim strengthen robustness of GCNs; for instance, training is used shield against malicious perturbations. these fail graphs which label scarcity a pressing issue. To overcome scarcity, self-training attempts iteratively assign pseudo-labels highly confident unlabeled nodes but such suffer serious degradation under graph this paper, we generalize noisy supervision as kind self-supervised learning method and then propose novel Bayesian self-supervision model, namely GraphSS, address Extensive experiments demonstrate GraphSS can not only affirmatively alert perturbations also effectively recover prediction classifier when These two advantages prove generalized over three classic across five public datasets.