SLFL: A Server-Less Federated Learning Framework in Software Defined UAV Networks

The unique characteristics of Unmanned Aerial Vehicle (UAV) networks, including limited resources, dynamic topology, and heterogeneity, pose significant challenges for integration with emerging technological scenarios such as Software Defined Network (SDN) and Federated Learning (FL) protocols. In this paper, we propose a consensus-based framework, called Server-Less FL (SLFL), specifically designed for FL in software-defined UAV networks. SLFL utilize real-time sensors and Global Positioning System (GPS) based Three-Dimensional (3D) location to leverage Reliable, Replicated, Redundant, And Fault-Tolerant (Raft) algorithm in distributed to ensure improved consensus among UAVs during the FL processing, allowing them to collaboratively train a global model while preserving data privacy, low overhead and minimizing communication disruptions. We performed model-based evaluation of SLFL and compared with existing solutions. Furthermore, SLFL effectively handles the challenges associated with UAV networks, including limited bandwidth, energy constraints, and mobility. The findings from this research can facilitate the adoption of FL in UAV applications. We concluded our research by highlighting pertinent future research directions for SLFL.