A forthcoming textbook from Springer, Federated Learning: From Theory to Practice by A. Jung, offers a hands-on introduction to building federated learning systems. The book targets students, engineers, and researchers who need to design scalable, privacy-preserving systems where data cannot or should not be centralized due to privacy, regulatory, or technical reasons.
The central contribution is a network-based view of federated learning. Devices are represented as nodes, and weighted edges capture both communication links and similarities between learning tasks. This structure lets each device train its own personalized model while still benefiting from collaboration with relevant peers. The training problem is formalized as generalized total variation minimization (GTVMin), ensuring that devices with similar tasks learn similar model parameters.
The author aims to keep the material accessible despite drawing on optimization theory and graph-based learning. The 252-page preprint, last revised in October 2026, is described as both mathematically principled and practically motivated, guiding readers through core ideas step by step. The final version is published by Springer Nature Singapore with ISBN 978-981-95-1008-5.