Optimization for Node Cooperation in Hierarchical Federated Learning
Xin Yuan Shen, Zhuo Li, Xin Chen · 2021
With the popularity of the Internet of things (IoT), a large number of data is generated in the network edge. The combination of edge computing and federated learning has become a key technology to reduce latency and energy consumption in mobile networks. The long delay is caused by the nodes transmitting parameters in Hierarchical Federated Learning (HFL). In this work, we propose a Dynamic Cooperative Cluster Algorithm (DCCA) for the delay minimization problem that is proved NP-hard. Moreover, communication between nodes adopts Device-to-Device (D2D) and opportunistic communication. In DCCA, clusters are generated according to the capabilities of different nodes for collaborative training models to reduce delay. We design the DCCA algorithm in two steps. For the first step, we propose an initial dynamic cooperative cluster algorithm based on similarity. For the second step, based on the computing capacity and transmission capacity of the nodes, another algorithm is proposed to adjust the dynamic cooperative cluster according to the core nodes. Extensive simulations are conducted to verify the performance of DCCA. We also observe a reduction in maximum delay by 12.44% and 18.70% respectively, compared with DOA and FedAvg.