Hyperbolic Graph Attention Network

Graph neural network (GNN) has shown superior performance in dealing with structured graphs, which attracted considerable research attention recently. Most of the existing GNNs are designed Euclidean spaces; however, real-world spatial data can be non-Euclidean surfaces (e.g., hyperbolic spaces). For example, biologists may inspect geometric shape a protein surface to determine its interaction other biomolecules for drug discovery. Although there is growing on generalizing surfaces, works these fields still scarce. In this paper, we exploit graph learn robust node representations graphs spaces. As gyrovector space framework provides an elegant algebraic formalism geometry, utilize Specifically, first use operations defined transform features graph; and proximity product spaces model multi-head mechanism setting; afterward, further devise parallel strategy using logarithmic exponential maps improve efficiency our proposed model. The comprehensive experimental results demonstrate effectiveness model, compared state-of-the-art methods.