Neural Networks on Hypergraph

Abstract With the development of deep learning on high-order correlations, hypergraph neural networks have received much attention in recent years. Generally, can be divided into two categories, including spectral-based methods and spatial-based methods. For methods, convolution operation is formulated spectral domain graph, we introduce typical (HGNN), with (Hyper-Atten), hyperbolic network (HHGNN), which extend computation to spaces beyond Euclidean space. defined groups spatially close vertices. We then present general (HGNN+) dynamic (DHGNN). Additionally, there are several that attempt reduce structure graph structure, so existing directly deployed. Lastly, analyze association comparison between areas described above (spectral-based, spatial-based), further demonstrating ability advantages constructing computing higher-order correlations data.