Federated Learning for IoT/Edge/Fog Computing Systems
Balqees Talal Hasan, Ali Kadhum Idrees · Apple Academic Press eBooks · 2024
With the help of a new architecture called Edge/Fog (E/F) computing, cloud computing services can now be extended closer to data generator devices. E/F computing, in combination with Deep Learning (DL), is a promising technique that is widely applied in numerous fields. In conventional DL architectures with E/F computing, data producers can train their models by repeatedly transmitting and communicating data with third-party servers, such as Edge/Fog or cloud servers. However, this architecture is often impractical due to extensive bandwidth needs, legal issues, and privacy risks. To address these challenges, a centralized server can be used to co-train the models through Federated Learning (FL) with distributed clients, including cars, hospitals, and mobile phones, while preserving data localization. FL facilitates group learning and model optimization, making it a motivating element in the E/F computing paradigm. Although previous studies have considered FL applications in E/F computing environments, the execution and hurdles of FL in the E/F computing framework have not been thoroughly covered. In order to identify advanced solutions, this chapter will provide a review of the application of FL in E/F computing systems. By conducting this study, researchers can gain a better understanding of how E/F computing 48 functions and how FL enables related concepts and technologies. Several case studies on the implementation of federated learning in E/F computing are currently being investigated. The open issues and future research directions will also be introduced.