Robust Federated Learning under Statistical Heterogeneity

Description

This book is dedicated to computer students specializing in computer sciences majoring in artificial intelligence, more precisely in federated learning frameworks. The growth of IoT has created massive, distributed, and privacy-sensitive datasets that centralized ML cannot handle. Federated Learning (FL) solves this by training shared models across devices without centralizing raw data. This book is split into two parts: the first covers foundational concepts: IoT architecture, ML/DL principles, FL protocols, and swarm/evolutionary optimisation methods; the second reviews the state of the art in robust FL across four challenges: statistical heterogeneity, client selection, model clustering, and security. A comparative analysis then identifies key gaps in the literature: a trade-off between robustness to heterogeneous data and resilience to attacks and underuse of hierarchical architectures.