Deep Reinforcement Learning for Mobile Traffic Offloading

In 5G networks, heterogeneous networks (HetNets) with small cells is a promising architecture to be deployed. The traffic offloading among macro and small cells is inevitably a key issue. Based on the cloud controlling structure, it is possible to design proactive strategies, so operation issues can be predicted and treated before suffering performance degradation. At the same time, the much more complex nature of 5G resource management is happen to be a suitable target to apply advanced machine learning approaches. In the project, we propose to apply deep reinforcement learning (DRL) on energy-efficient mobile traffic offloading. Taking advantage of our traffic forecasting works, we will investigate the DRL model for 5G networking issues, and further provide suggestion for future 5G resource management works.