AI glossary Data & Retrieval
What is federated learning?
← All glossary termsA way to train a shared model across many devices or organisations, keeping raw data where it is and sharing only model updates.
Explained
Federated learning trains a machine learning model across many separate data holders, such as phones, hospitals or bank branches, without collecting their raw data in one place. Each participant trains the current model on its own data and sends back only the resulting update; a coordinating server combines the updates into a new shared model and repeats the process. The approach was introduced by Google researchers, whose federated averaging algorithm remains the standard reference. It suits regulated and privacy-sensitive settings where data cannot or should not leave its source.
How it works
- A coordinator sends the current model to the participants.
- Each participant trains it for a while on its own local data.
- Participants send back model updates, not the data itself.
- The coordinator averages or otherwise combines the updates into an improved model.
- The cycle repeats until the model is good enough.
Where it is used
On devices, it lets a keyboard or voice model improve from how people use their phones without uploading what they type or say. Across organisations, it lets hospitals, banks or insurers build a model from their combined experience while each keeps its records in-house.
It is also used inside a single company when data sits in different countries or business units with rules that stop it being centralised.
Limits to know
- Updates can still leak information about the data, so federated learning is usually combined with secure aggregation, differential privacy or both.
- Data differs a lot between participants, which makes training slower and less stable than with centralised data.
- Communication and coordination are the main costs, especially with many devices.
- Participants need enough data and compute, and agreement on governance: who owns the model, who can use it and how quality is checked.
Frequently asked questions
What is federated learning in simple terms?
Instead of bringing the data to the model, you take the model to the data. Each participant improves the model locally and only the improvements are shared and combined.
Is federated learning secure and private?
It reduces risk because raw data stays at its source, but it is not private by itself: model updates can reveal information. It is usually paired with techniques such as secure aggregation and differential privacy, plus normal security controls.
What is an example of federated learning?
Phone keyboards that learn better word suggestions from many users' typing, without the typed text leaving each phone. In healthcare, several hospitals training a diagnostic model together without sharing patient records.
Does federated learning help with GDPR?
It can help with data minimisation, because personal data is not centralised. It does not remove obligations: updates may still count as personal data in some cases, and a legal and privacy assessment is still needed.
What is the difference between federated learning and distributed training?
Distributed training splits a job across machines in one data centre, with data the organisation controls and can move freely. Federated learning works across participants who keep their own data, often on unreliable devices or across organisational boundaries.
Where it shows up at Thinkia
Further reading
Related terms
Sources
- McMahan et al. · Communication-efficient learning of deep networks from decentralized data (arXiv)
- Kairouz et al. · Advances and open problems in federated learning (arXiv)
- Google Research · Federated learning: collaborative machine learning without centralized training data
Updated: 2026-10-21