A new federated learning framework named FedCCS improves convergence efficiency and predictive performance across heterogeneous computing environments by combining model representation clustering with straggler-aware client selection, researchers reported in the International Journal of Machine Learning and Cybernetics.
In distributed federated learning systems, collaborative model training takes place across separate clients without directly exposing private user data. However, the authors noted that training efficiency is frequently compromised by statistical non-IID data distributions and system heterogeneity, which lead to gradient divergence among models and processing delays caused by straggler devices.
To mitigate skew in data distribution, the FedCCS framework extracts compact representations from the weights of clients’ linear layers. The algorithm then clusters these participating clients using Affinity Propagation based on the similarity of their model representations.
Within each resulting cluster, FedCCS estimates local device training times and selects only those clients projected to complete their workloads prior to an optimal cut-off threshold. This adaptive latency threshold mechanism is designed to maximize overall throughput while bypassing slower hardware.
The researchers implemented FedCCS using PyTorch and RabbitMQ communication, evaluating the architecture against representative baselines on benchmark datasets including CIFAR-10, DOMAIN, and CIC-IDS2017. Experimental results demonstrated consistent improvements in time-to-accuracy alongside competitive or superior predictive accuracy under both IID and non-IID conditions, with several configurations showing substantial reductions in convergence time and energy consumption.
Subject of Research: Heterogeneous federated learning optimization
Article Title: FedCCS: joint client clustering and selection for federated learning under statistical and system heterogeneity
Article References: Viet Duc, M., Thu Phuong, T., Minh Dang, P., Hai, V., & Huu Thanh, N. (2026). FedCCS: joint client clustering and selection for federated learning under statistical and system heterogeneity. International Journal of Machine Learning and Cybernetics, 17(11), Article 528. https://doi.org/10.1007/s13042-026-03352-6
Image Credits: AI Generated
DOI: 10.1007/s13042-026-03352-6
Keywords: federated learning, client clustering, straggler mitigation, Affinity Propagation, system heterogeneity, distributed training, machine learning, FedCCS, joint, client, clustering, selection
Cite Scienmag News
Veronica Carney. (November 9, 2026). Clustering and Latency Selection Accelerate Heterogeneous Federated Learning. Scienmag. https://scienmag.com/clustering-and-latency-selection-accelerate-heterogeneous-federated-learning/
Veronica Carney. "Clustering and Latency Selection Accelerate Heterogeneous Federated Learning." Scienmag, 9 November 2026, https://scienmag.com/clustering-and-latency-selection-accelerate-heterogeneous-federated-learning/. Accessed 20 November 2026.
Veronica Carney. "Clustering and Latency Selection Accelerate Heterogeneous Federated Learning." Scienmag. November 9, 2026. https://scienmag.com/clustering-and-latency-selection-accelerate-heterogeneous-federated-learning/








