Survey of Hypergraph Neural Networks and Its Application to Action Recognition.

With the development of deep learning, graph neural networks have attracted ever-increasing attention due to their exciting results on handling data from non-Euclidean space in recent years. However, existing graph neural networks frameworks are designed based on simple graphs, which limits their ability to handle data with complex correlations. Therefore, in some special cases, especially when the data have interdependence, the complexity of the data poses a significant challenge to traditional graph neural networks algorithm. To overcome this challenge, researchers model the complex relationship of data by constructing hypergraph, and use hypergraph neural networks to learn the complex relationship within data, so as to effectively obtain higher-order feature representations of data. In this paper, we first review the basics of hypergraph, then provide a detailed analysis and comparison of some recently proposed hypergraph neural networks algorithm, next some applications of hypergraph neural networks for action recognition are listed, and finally propose potential future research directions of hypergraph neural networks to provide ideas for subsequent research.