How does federated learning reduce data sharing?

Federated learning trains a shared machine-learning model across separate devices or organisations while keeping the underlying training records in their local environments.

A typical process involves:

• Sending a model to participating devices or organisations.

• Training it locally using their own information.

• Returning model updates for combination into an improved shared model.

This can reduce the need to collect everyone’s raw records in one central database. However, model updates can still reveal information about the data used to produce them.

Protections may therefore include secure aggregation, restrictions on participation and differential privacy. The appropriate combination depends on who can inspect updates and what attacks the system needs to withstand.

The ICO’s federated learning guidance explains these limitations. Keeping records locally does not, by itself, prove that the process is anonymous or remove obligations associated with processing personal data.