Search papers, labs, and topics across Lattice.
This paper extends the Flower federated learning framework by introducing a Semi-Asynchronous Federated Learning (SAFL) strategy that addresses client heterogeneity and straggler effects. The FedSaSync method allows for partial synchronization among clients, significantly improving robustness and reducing idle time compared to traditional fully synchronous approaches. Experimental evaluations demonstrate that this semi-asynchronous approach effectively balances convergence stability and system efficiency in heterogeneous edge and distributed learning environments.
Partial synchronization in federated learning can drastically cut idle time and enhance robustness in heterogeneous client environments.
This paper presents an extension of the Flower federated learning framework to support Semi-Asynchronous Federated Learning. The proposed approach adapts the traditional synchronous paradigm to better handle client heterogeneity and straggler effects. By introducing a semi-asynchronous training strategy, the system allows partial synchronization among clients while maintaining training efficiency and scalability. We implement and evaluate the proposed modification within Flower, instantiated as the FedSaSync strategy, demonstrating improved robustness and reduced idle time compared to fully synchronous baselines in heterogeneous environments. The results show that SAFL can balance convergence stability and system efficiency in heterogeneous environments typical of edge and distributed learning scenarios.
