Federated learning

Federated learning (FL) trains a shared model across many clients — phones, hospitals, banks — without collecting their raw data. Data stays where it was created; only model updates travel.

A typical round:

• The server sends the current global model to a sample of clients.
• Each client trains locally on its own data for a few epochs.
• Clients send back their updated weights (or weight changes).
• The server aggregates them into a new global model.

It enables learning from data that is private, regulated or too costly to move. Real deployments include next-word prediction on phone keyboards and collaborative medical-imaging models across hospitals.