(A) study on federated learning in wireless communication systems

Abstract

In this dissertation, we studied the federated learning in the wireless communication systems. In the federated learning, there are a single parameter server and multiple distributed devices collaboratively trains a model by sharing their updated local models. When applying the federated learning in the wireless communication systems, various schemes are studied to improve the training performance of it. Firstly, the device grouping based over-the-air computation federated learning scheme is proposed to prevent malicious attacks of the Byzantine devices. Also, in the orthogonal communication system, the communication-efficient federated learning scheme is studied by reducing the communication overhead via projection based compression. Finally, in the over-the-air computation based federated learning, the differential privacy preserving federated learning is proposed by using the inherent property of over-the-air computation.