This chapter explores learning paradigms in cross-device federated recommendation systems, analyzing statistical machine learning, deep learning, meta-learning, and reinforcement learning (RL). It reviews key studies, highlighting models, techniques, and datasets used in FedRec. The chapter addresses challenges such as data heterogeneity, communication efficiency, and model adaptability. It emphasizes the importance of efficient aggregation strategies and advanced privacy-preserving techniques to handle diverse data and models. A comparative analysis of different approaches reveals their strengths and limitations, offering insights to advance federated recommendation systems. Finally, the chapter discusses future research directions to enhance system scalability and effectiveness.