Federated Learning

Today, machine learning (ML) sees a variety of applications in business processes, whether toward individuals or other companies. However, data privacy often comes short when vast amounts of data are to be transferred from clients participating in training processes to the ML service provider. federated learning (FL) seeks to address this issue by moving the model to the data, ensuring that data does not leave end devices. Depending on different use cases and industries, we can observe a variety of architectural patterns and algorithms to assist with the inherent requirements of FL in averaging results and ensuring stable communication. Nonetheless, maintaining privacy impels new challenges concerning algorithmic performance and fairness among participating clients. FL has the potential to revolutionize the way we train machine learning models, enabling collaboration across different devices and organizations without compromising data privacy and security. In this chapter, we portray the basic concepts of FL, discuss common architectures, address inherent challenges, and shed light on fairness aspects in conjunction with performance and privacy. We conclude by sharing considerations for implementing FL processes in practice and giving a short outlook on the potentials of FL from practice and research standpoints.