Source: Xu, Keyulu, et al. "Representation learning on graphs with jumping knowledge networks." arXiv preprint arXiv:1806.03536 (2018). Problem to be addressed There are two major problems to be addressed, deep network degradation and the bias from graph structure. Degradation as network goes deeper Although recent developments of graphic neural network have achieved state-of-the-art results on several studies with graph structures. It is concerned that most of then faces the same challenge as the CNN had, i.e. the performance degrades as network grows deeper. Many of the recent approaches broadly follow a neighborhood aggregation (or "message passing" scheme), which have been shown to generalize the Weisfeiler-Lehman graph isomorphism test. Yet, such aggregation schemes sometimes lead to surprises. For example, it has been observed that the best performance with one of the state-of-the-art models, Graph Convolutional Networks (GCN), is ach...