Minimizing latency in cognitive UAV-aided edge networks using partial federated learning [1]

Abstract

Unmanned aerial vehicles (UAVs) are increasingly deployed as aerial relays to support Internet of Things (IoT) applications in digital cities where low-latency communication is essential. In addition, UAVs can serve as aerial computing nodes and enable edge computing for IoT devices in remote locations where conventional infrastructure is unavailable. However, despite these benefits, limited communication resources, such as bandwidth and energy constraints of IoT devices, pose a significant challenge to the implementation of federated learning (FL) in UAV-assisted networks. FL is a distributed machine-learning approach that allows devices to collaboratively train a model without the need to share local data. To address these challenges, this paper proposes a partial FL framework for UAV-assisted edge computing in a cognitive radio network. The proposed method jointly optimizes the transmit and compute powers of cognitive IoT devices, cognitive UAVs, and base stations. These cognitive IoT devices dynamically adapt their transmission parameters (such as frequency and power) based on real-time wireless spectrum conditions to optimize communication efficiency and reduce interference while adhering to the transmit-power limitations of the secondary network. We provide an iterative algorithm using the Davidon-Fletcher-Powell approach to solve this non-convex problem. The simulation results demonstrate that the proposed partial FL system can significantly reduce latency by approximately 12.81 % compared to the conventional FL approach. Thus, the proposed method can play a crucial role in enhancing low-latency edge computing services for IoT devices in remote locations.