This handbook aims to serve as a one-stop, reliable resource, including curated surveys and expository contributions on federated learning. It covers a comprehensive range of topics, providing the reader with technical and non-technical fundamentals, applications, and extensive details of various topics. The readership spans from researchers and academics to practitioners who are deeply engaged or are starting to venture into the realms of trustworthy federated learning. First introduced in 2016, federated learning allows devices to collaboratively learn a shared model while keeping raw data localized, thus promising to protect data privacy. Since its introduction, federated learning has undergone several evolutions. Most importantly, its evolution is in response to the growing recognition that its promise of collaborative learning is inseparable from the imperatives of privacy preservation and model security.

The resource is divided into four parts. Part 1 (Security and Privacy) explores the robust defense mechanisms against targeted attacks and addresses fairness concerns, providing a multifaceted foundation for securing Federated Learning systems against evolving threats. Part 2 (Bilevel Optimization) unravels the intricacies of optimizing performance in federated settings. Part 3 (Graph and Large Language Models) addresses the challenges in training Graph Neural Networks and ensuring privacy in Federated Learning of natural language models. Part 4 (Edge Intelligence and Applications) demonstrates how Federated Learning can empower mobile applications and preserve privacy with synthetic data.

This handbook aims to serve as a one-stop, reliable resource, including curated surveys and expository contributions on federated learning. It covers a comprehensive range of topics, providing the reader with technical and non-technical fundamentals, applications, and extensive details of various topics. The readership spans from researchers and academics to practitioners who are deeply engaged or are starting to venture into the realms of trustworthy federated learning. First introduced in 2016, federated learning allows devices to collaboratively learn a shared model while keeping raw data localized, thus promising to protect data privacy. Since its introduction, federated learning has undergone several evolutions. Most importantly, its evolution is in response to the growing recognition that its promise of collaborative learning is inseparable from the imperatives of privacy preservation and model security.

The resource is divided into four parts. Part 1 (Security and Privacy) explores the robust defense mechanisms against targeted attacks and addresses fairness concerns, providing a multifaceted foundation for securing Federated Learning systems against evolving threats. Part 2 (Bilevel Optimization) unravels the intricacies of optimizing performance in federated settings. Part 3 (Graph and Large Language Models) addresses the challenges in training Graph Neural Networks and ensuring privacy in Federated Learning of natural language models. Part 4 (Edge Intelligence and Applications) demonstrates how Federated Learning can empower mobile applications and preserve privacy with synthetic data.