Hierarchical Semi-Asynchronous Federated Learning Based on Over-the-Air Computation
Jishi Jiang, Shijie Shi, Yitong Li, Fasong Wang, Yanbin Zhang · 2023
With the explosive growth in demand from application scenarios, such as industrial IoT, the clients process data locally are increasingly imminent. However, due to the uncertainty of the local network, some dropped clients may result in a slower convergence speed and a worse test accuracy. To solve the above problems, we propose a new framework to alleviate the strict synchronization requirements of FedAvg algorithm with Over-the-Air Computation(OTC), as well as to reduce the consumption of communication resources. In our framework, we introduce free clients and corresponding deadline to reduce the latency caused by dropped clients. The free clients and deadline time are realistic set, and we also analyze their effectiveness. We found that under the premise of achieving the same test accuracy, the time consumption with free clients is 14% less than that without free clients.