Learning Data-Driven Propagation Mechanism for Graph Neural Network

A graph is a relational data structure suitable for representing non-Euclidean structured data. In recent years, neural networks (GNN) and their subsequent variants, which utilize deep to complete analysis representation, have shown excellent performance in various application fields. However, the propagation mechanism of existing methods relies on hand-designed GNN layer connection architecture, prone information redundancy over-smoothing problems. To alleviate this problem, we propose data-driven adaptively propagate between layers. Specifically, construct bi-level optimization objective use gradient descent algorithm learn forward improves efficiency learning different combinations multilayer networks. The experimental results model seven benchmark datasets demonstrate effectiveness proposed method. Furthermore, combining with models, such as Graph Attention Networks, can consistently improve these models.