Edge-Cloud Collaborative Computation Offloading for Federated Learning in Smart City.

With the continuous transformation and development of information technologies, the smart city is becoming a promising paradigm to deal with the enormous network data. Among them, federated learning technology has emerged as a key tool for intelligent analysis and data processing. It effectively guarantees the privacy and security of users by shifting conventional data storage and model training to local devices. Nevertheless, strong convergence performance necessitates numerous rounds of data exchange, which is uneconomical for edge platforms with limited resources. In view of this, we study the offloading of federated learning models in the edge-cloud collaborative smart city. Firstly, we transform FL models into a helpful structure for improving the efficient aggregation of servers. Then, to lower the energy and time cost of mobile devices throughout the model transmission and aggregation process while maintaining a high resource utilization level of edge servers, we develop an efficient computational offloading mechanism. Finally, the experimental results demonstrate the efficiency of our proposed method.