Can Routing Be Effectively Learned in Integrated Heterogeneous Networks?

With the advent of 5G and facing future 6G, various networks tend to be linked together to form an integrated heterogeneous network (Inte-HetNets). Inte-HetNets bring new challenges to routing due to the need of crossing multiple network domains. Traditional routing methods are formidable to effectively support routing in Inte-HetNets. Machine learning is regarded as an promising technology to achieve such a goal, which has attracted efforts of many researchers. However, the literature still lacks a review on current research advance. In this paper, we review existing intelligent routing schemes based on machine learning in Inte-HetNets. We first introduce mainstream machine learning methods applied into routing. Then, we provide a taxonomy of learning-empowered routing schemes in Inte-HetNets by classifying them into three types based on routing scenarios: routing in ad hoc networks, routing in fixed backbone networks, and routing across network domains. Subsequently, we propose a set of requirements on learning-empowered routing in Inte-HetNets and employ these requirements to review the current literature. Finally, we explore several open issues based on our review and indicate future research directions of intelligent routing in Inte-HetNets.