Cluster Based Pseudo Hierarchical Decentralized Federated Learning in UAV Networks

The technological advancements in Unmanned Aerial Vehicles (UAVs) have brought significant changes in various domains including surveillance, agriculture and disaster rescue. The convergence of Machine Learning (ML) and UAV networks contributes significantly to their automation and decision-making capabilities. Traditional ML techniques are centralized, i.e., they face many issues such as privacy due to data sharing, scalability and single-point failure. In this work, we propose a hierarchical decentralized framework for Federated Learning (FL) that addresses all the aforementioned issues. The proposed framework, Cluster Based Pseudo Hierarchical Decentralized Federated Learning (PHDFL), is tailored to UAV networks for learning where the learning and aggregation tasks are distributed among different UAVs in the network. This also introduces the concept of pseudo-hierarchy as all the UAVs are at the same level due to Decentralized Federated Learning (DFL) but the learning happens in a hierarchical manner where the network is divided into clusters and each cluster has a cluster head which in then communicates with other cluster heads. The effectiveness of the proposed framework is evaluated through experiments in terms of learning time, energy consumed and convergence of the model.