Reliable Graph Neural Networks via Robust Aggregation

This work is motivated by the observation that adversarially injected edges effectively can be viewed as additional samples to a node's neighborhood aggregation