Intelligent hierarchical federated learning system based on semi-asynchronous and scheduled synchronous control strategies in satellite network

• DOI: 10.1007/s43684-025-00095-z
• Publication Date: 2025-03-20T21:46:29Z
• Authors: Qiang Mei; Rui Huang; Duo Li; Jingyi Li; Nan Shi; Mei Du; Yingkang Zhong; Chunqi Tian
• Keywords: Asynchronous learning

Abstract Federated learning (FL) is a technology that allows multiple devices to collaboratively train global model without sharing original data, which hot topic in distributed intelligent systems. Combined with satellite network, FL can overcome the geographical limitation and achieve broader applications. However, it also faces issues such as straggler effect, unreliable network environments non-independent identically (Non-IID) samples. To address these problems, we propose an hierarchical system based on semi-asynchronous scheduled synchronous control strategies cloud-edge-client structure for network. Our effectively handles client requests by distributing communication load of central cloud various edge clouds. Additionally, server selection algorithm edge-client strategy minimize clients’ waiting time, improving overall efficiency process. Furthermore, center-edge ensures timeliness partial models. Based experiment results, our proposed demonstrates distinct advantage accuracy over traditional FedAvg, achieving 2% higher within same time frame reducing 52% training target accuracy.