Federated learning trains models across distributed data that never leaves its source: devices or institutions compute local updates, and only the updates, not
Training models across distributed data that never leaves its source: devices or institutions compute local updates, and only the updates, not the raw data, are aggregated into a shared model.
When data cannot move: hospitals jointly training clinical models, phones improving keyboards locally, or banks sharing fraud signal without sharing customers. It is often paired with differential privacy and secure aggregation for end-to-end guarantees.