Machine learning and its application in communication networks

Future-generation wireless networks (5G and beyond) must accommodate the surging growth of mobile data traffic and support a high density of mobile users with a variety of services and applications. With the dynamics and uncertainty inherently existing in the wireless network environments, conventional approaches of service and resource management that require complete and perfect knowledge of the systems become inefficient or even inapplicable. Deep reinforcement learning (DRL) based machine learning (ML) approaches allow network entities to learn and build knowledge about the networks to make optimal decisions locally and independently. Due to user mobility and network dynamics, the optimal policies learned from DRL need to be adaptive to the new environment. In this case, the transfer learning approach can be used to leverage learned experiences from a source environment to facilitate and speed up the learning process in the new environment. Meanwhile, in 5G and beyond, the massive connectivity (e.g. IoT) and the computing capability on edge sides are of great protrusion. To achieve low latency, preserve user data privacy, and reduce the burden of network, real edge artificial intelligence is important, which demands a decentralized ML model. Federated learning (FL) provides the paradigm to train ML model in a decentralized manner. In this talk, I will introduce these three machine learning techniques (i.e., deep reinforcement learning, transfer learning, and federated learning) and their applications in communication networks.