Federated Learning is a distributed machine learning method where multiple clients train models locally and only transmit their parameters or gradients to a central server.
The server aggregates these updates, usually by weighted averaging (e.g., FedAvg), and distributes the updated global model back to the clients.
This approach enables training models on distributed, private datasets without direct data sharing.
Challenges include heterogeneous data distribution, varying computational capacities of clients, as well as communication costs.
Federated Learning is often used in areas such as mobile devices, healthcare, or IoT, where data privacy is important.