Personalized federated learning: theory and open problems

Federated Learning (FL) is a distributed and privacy-preserving machine learning technique in which a group of clients collaborates with a server to learn a global model without sharing clients' data. One challenge with FL is statistical diversity among clients, which restricts the global model from delivering good performance on each client's task. A common approach to address this challenge is to find a “personalized model” that is stylized for each client's data. This chapter introduces and reviews several current personalized federated learning (pFL) methods to address this challenge. To give more insights, state-of-the-art pFL algorithms are then discussed and compared with detailed experiments. Finally, open problems are highlighted for potential future works.