Federated Flowchart: Overview of State-of-the-Arts based on Federated Learning Process

Abstract

Federated learning centrally trains global statistical models by aggregating the local models trained on each device with localized data. Accordingly, federated learning protects the privacy of each device by eliminating the need to transmit distributed data generated in real-time on each device to a central data center. However, unlike other machine learning paradigms, this new paradigm raises new problems in large-scale and high-density networks composed of various devices, such as privacy protection due to the numerous devices, and distributed optimization due to the different resources of each device. In this paper, we describe the characteristics, challenges, and process of federated learning, and provide an overview of state-of-the-arts according to each phase of the process. © 2022 IEEE.