← Back to glossary
🧠

Federated Learning

Privacy

Differential Privacy AI Data Privacy Data Minimization EU AI Act

Distributed ML training where clients compute updates locally and share aggregated model parameters.

Definition

Federated learning performs training across many clients; each client computes updates locally and shares aggregated updates to a coordinator. Privacy often requires secure aggregation and/or differential privacy.

In plain English Distributed ML training where clients compute updates locally and share aggregated model parameters.

Why this matters

Why it matters: Without safeguards, gradients or updates can leak sensitive signals; governance and threat modeling are essential.

Example

Example: Use secure aggregation so the server cannot see individual updates and apply differential privacy to reduce leakage risk.