XPPFEL

Abstract

Intelligent decision-making for autonomous flying drones and vehicles requires real-time processing capabilities, while at the same time protecting the privacy of collected data, reducing latency, and guaranteeing effective communication. Centralized machine learning frameworks not only leak sensitive data but also result in additional communication costs in changing edge network conditions. In this paper, we present a novel framework called Explainable and Privacy-Preserving Federated Edge Learning (XPP-FEL) aimed towards intelligent autonomous unmanned aerial vehicles (UAVs) and vehicle-based systems. The presented framework leverages federated learning with edge computing principles to train the models in a decentralized manner while avoiding transmitting the raw sensor data. Differential privacy and secure aggregation algorithms have been applied in the framework to protect the trained models from any potential data leakage or cyber attacks. For improving the framework's transparency and interpretability, explainable AI methodologies such as SHAP and LIME have been employed.