Behave Differently when Clustering: A Semi-asynchronous Federated Learning Approach for IoT

• DOI: 10.1145/3639825
• Publication Date: 2024-01-25T12:41:27Z
• Authors: Boyu Fan; Xiang Su; Sasu Tarkoma; Pan Hui
• Keywords: Smart sensing; Computer and information sciences; Federated learning; 0202 electrical engineering, electronic engineering, information engineering; Deep learning; Edge computing; 02 engineering and technology

The Internet of Things (IoT) has revolutionized the connectivity diverse sensing devices, generating an enormous volume data. However, applying machine learning algorithms to devices presents substantial challenges due resource constraints and privacy concerns. Federated (FL) emerges as a promising solution allowing for training models in distributed manner while preserving data on client devices. We contribute SAFI , semi-asynchronous FL approach based clustering achieve novel in-cluster synchronous out-cluster asynchronous mode. Specifically, we propose three-tier architecture enable IoT processing edge design selection module effectively group heterogeneous their capacities. performance been extensively evaluated through experiments conducted real-world testbed. As heterogeneity increases, surpasses baselines terms convergence time, achieving speedup approximately × 3 when ratio is 7:1. Moreover, demonstrates favorable non-independent identically settings requires lower communication cost compared FedAsync. Notably, first Java-implemented holds significant promise serve efficient algorithm environments.