Over the past decades, machine learning techniques have demonstrated excellent superiorities in a wide range of fields, such as computer vision, natural language processing, etc. Through efficient utilization of a huge amount of data, machine learning techniques can solve problems that are hard or impossible for conventional model-based solutions, because the simplified models cannot effectively approximate actual scenarios while complicated models cannot be practically solved in a mathematically rigorous sense. In the meantime, future wireless communication systems are becoming increasingly complex due to diverse practical demands and communication applications. This makes it urgent to find alternatives to conventional solutions and warrants a paradigm shift towards the machine learning-driven direction. Although the convergence of wireless communication and machine learning is just unfolding, it has already achieved initial success in academic research and practical applications. This paper reviews the latest research of machine learning in wireless communications. We highlight key technologies of machine learning-driven signal processing, end-to-end communications and semantic communications, machine learning-based resource allocation, and federated learning of distributed systems. Furthermore, open challenges and potential opportunities in the convergence of machine learning and wireless communication are also illustrated.