Federated learning
Also known as: federated learning
Federated learning trains a model across many devices (phones, servers) without gathering all the raw data in one place.
- Each device trains locally, sends back updates, not data.
- Raw data stays on the device.
- Often combined with differential privacy.
The privacy story: federated learning flips the model — instead of data coming to the model, the model goes to the data. That's what keeps raw data private.