Abstract
Due to the sensitivity of data, federated learning (FL) is employed to enabledistributed machine learning while safeguarding data privacy and accommodatingthe requirements of various devices. However, in the context ofsemi-decentralized federated learning (SD-FL), clients' communication andtraining states are dynamic. This variability arises from local trainingfluctuations, heterogeneous data distributions, and intermittent clientparticipation. Most existing studies primarily focus on stable client states,neglecting the dynamic challenges present in real-world scenarios. To tacklethis issue, we propose a trust-aware client scheduling mechanism (TRAIL) thatassesses client states and contributions, enhancing model training efficiencythrough selective client participation. Our focus is on a semi-decentralizedfederated learning framework where edge servers and clients train a sharedglobal model using unreliable intra-cluster model aggregation and inter-clustermodel consensus. First, we develop an adaptive hidden semi-Markov model (AHSMM)to estimate clients' communication states and contributions. Next, we address aclient-server association optimization problem to minimize global trainingloss. Using convergence analysis, we propose a greedy client schedulingalgorithm. Finally, our experiments conducted on real-world datasetsdemonstrate that TRAIL outperforms state-of-the-art baselines, achieving animprovement of 8.7\% in test accuracy and a reduction of 15.3\% in trainingloss.