Responsible Tech MaxxingIntermediatePrivacy90 sec

Federated learning

Federated learning trains a shared model across many devices without any individual device's raw data ever leaving that device, sharing only model updates instead.

Mark as completed

Prerequisite: Differential privacy

What matters

Because raw personal data never has to be centralized, federated learning can reduce privacy risk while still allowing a useful shared model to be trained across many users.

Example

A keyboard app might improve its predictive text model by learning from typing patterns on each phone locally, sending only aggregated model updates back, never the actual typed text.

Common misconception

Federated learning does not automatically guarantee complete privacy; the shared model updates themselves can sometimes still leak information about the underlying data.

Recall question

What does federated learning share between devices, instead of raw personal data?

Keywords and related concepts