This chapter explores major variants of federated learning (FL) that arise from different structures of local data and network connectivity. It shows how each variant can be seen as a special case of the GTVMin framework introduced earlier. The chapter begins with single-model FL, where all devices collaboratively train a shared model. Next, it covers clustered FL, which groups devices into clusters based on data similarity, allowing each cluster to train its own model. Horizontal FL is discussed as a setup where devices hold different samples of a shared dataset, while vertical FL covers cases where devices share the same data points but observe different features. The chapter then introduces personalized FL, where only parts of a model are shared across devices, allowing customization for local data. Finally, it presents few-shot learning as a setting where FL helps train models across categories with limited data. Each variant reflects different assumptions about data distribution and model structure. The chapter explains how to implement these variants using the same core optimization tools, making GTVMin a flexible framework for diverse FL applications.