Can a model learn without collecting all the data in one place?
Federated learning lets several devices or organisations train a model together without necessarily pooling all their raw data.
What is it?
In federated learning a model is trained across many devices or organisations without collecting all the raw data in one place. Each device sends only updates to the model, which are then combined. That lets the data stay closer to their owner.
An everyday example
Your phone keyboard learns which words you use often and sends only a summary of changes to the shared model, not your messages.
Why does it matter?
It lets hospitals, banks or devices cooperate when they may not or do not want to share data, and reduces the risk of large collections of personal data.
A common misconception
That federated learning guarantees privacy on its own. Data can sometimes be inferred from model updates, so it needs extra protection.
In more detail
How does it work?
Devices do part of the training locally and send updates to a shared process.
Does that guarantee privacy?
Not automatically. The updates can reveal information, so extra safeguards are needed.
When is it useful?
When it makes no sense to keep data in one place, for privacy or practical reasons.
What is the downside?
Different devices, slow connections and uneven data make coordinated training harder.
Remember: The data may stay local, but the risks do not disappear.
Reliable sources
Last reviewed 1 November 2026. An explainer, not professional, legal or medical advice. For rules and figures that change, check the official version.