graph neural networks (GNNs) have significantly improved the representation
power for graph-structured data. Despite of the recent success of GNNs, the
graph convolution in most GNNs have two limitations. Since the graph
convolution is performed in a small local neighborhood on the inp