The burgeoning advances in machine learning and wireless technologies are forging a new paradigm for future networks, which are expected to possess higher degrees of intelligence via the inference from vast dataset and being able to respond to local events in a timely manner. Due to the sheer volume of data generated by end-user devices, as well as the increasing concerns about sharing private information, a new branch of machine learning models, namely federated learning, has emerged from the intersection of artificial intelligence and edge computing. In contrast to conventional machine learning methods, federated learning brings the models directly to the device for training, where only the resultant parameters shall be sent to the edge servers. The local copies of the model on the devices bring along great advantages of eliminating network latency and preserving data privacy. Nevertheless, to make federated learning possible, one needs to tackle new challenges that require a fundamental departure from standard methods designed for distributed optimizations. In this paper, we aim to deliver a comprehensive introduction of federated learning. Specifically, we first survey the basis of federated learning, including its learning structure and the distinct features from conventional machine learning models. We then enumerate several critical issues associated with the deployment of federated learning in a wireless network, and show why and how technologies should be jointly integrated to facilitate the full implementation from different perspectives, ranging from algorithmic design, on-device training, to communication resource management. Finally, we conclude by shedding light on some potential applications and future trends.
