Most real-world networks are noisy and incomplete samples from an unknown
target distribution. Refining them by correcting corruptions or inferring
unobserved regions typically improves downstream performance. Inspired by the
impressive generative capabilities that have been used to co
GND-Nets, a new graph neural network that exploits local and global neighborhood information, is proposed to mitigate the over-smoothing and under-smoothing problems of Graph Convolutional Networks, using a new graph diffusion method called neural diffusions, which integrate neural networks into the conventional linear and nonlinear graph diffusions.