A contemporary survey of recent advances in federated learning: Taxonomies, applications, and challenges

Abstract

The Internet of Things (IoT) has embedded itself in our daily lives, offering smart services and AI-driven applications. However, traditional AI methods face challenges due to centralized data management in vast IoT networks and privacy concerns. Federated Learning (FL) solves this by training AI directly on distributed IoT devices, avoiding extensive data sharing. This article provides interested researchers of FL with an in-depth and up-to-date exploration of the multifaceted of FL in a privacy-centric era. This study is structured with three principal objectives. Firstly, within the realm of taxonomies for FL, the study aims to identify and categorize the existing frameworks utilized in FL research. Furthermore, it seeks to provide a thorough overview of how FL is delineated and conceptualized in current literature. Secondly, the study endeavors to explore the applications of FL by scrutinizing and classifying various use cases across diverse domains and industries. This objective aims to illuminate the applications and industries benefitting from FL implementations. Thirdly, the study addresses the inherent challenges in FL by examining the associated limitations. The study strives to offer insights into the technical and practical obstacles encountered in FL scenarios, contributing to a comprehensive understanding of the challenges within this dynamic field. This survey serves as a roadmap towards establishing sustainable massive IoT networks. Additionally, it paves the way for interested researchers to pursue and advance the vision of FL's evolving role in IoT.