Towards Deeper Graph Neural Networks with Differentiable Group Normalization

Several attempts have been made to tackle the issue by bringing linked node pairs close and unlinked pairs distinct. However, they often ignore the intrinsic community structures and would result in sub-optimal performance.