FEDERATED LEARNING AND ITS APPLICATIONS IN MULTI-UAVS SYSTEM

This dissertation explores Federated Learning (FL), a novel distributed machine learning paradigm with significant potential in many practical applications, in both algorithmic and application aspects. The primary contribution of this work is to examine existing challenges and solutions within the FL framework and its potential applications in multi-UAV systems. The dissertation is structured into eight chapters. Chapter 1 provides a general introduction to FL and its core challenges. Chapter 2 presents a literature review of recent studies addressing various FL challenges. Chapter 3 outlines the motivations and key contributions of this dissertation. Chapters 4 through 7 focus on the main research efforts. Specifically, Chapters 4 and 5 address two major challenges, class imbalance and domain shift, by proposing new algorithmic solutions. Chapter 6 introduces a real-time urban object detection system for UAVs leveraging FL framework. Chapter 7 investigates the causes of accuracy degradation in FL-based UAV systems and applies novel techniques to enhance performance. Finally, Chapter 8 summarizes the key findings and contributions of this dissertation and suggests future research directions in FL and its applications in multi-UAV systems.