Graph Neural Networks for Communication Networks: A Survey

Abstract

Communication networks are an important infrastructure in contemporary society. In recent years, based on the advancement and application of machine learning and deep learning in communication networks, the most advanced deep learning method, Graph Neural Network (GNN), has been applied to understand multi-scale deep correlations, provide generalization ability, and improve the accuracy indicators of predictive modeling. In this survey, we reviewed various issues using different graph based deep learning models in different types of communication networks. Optimize control strategies, including offloading strategies, routing optimization, resource allocation, etc. Finally, we discussed potential research challenges and future directions.