Federated Learning is a technique for training a shared AI model across many devices while keeping each device’s data private and local. The data never leaves the device; only model updates are shared.
What it means in plain English
Normally, training a model means gathering all the data in one place — a privacy risk. Federated learning flips this: the model is sent to the devices, each device trains it a little on its own local data, and only the resulting model improvements (not the raw data) are sent back and combined. This lets a model learn from everyone’s data without anyone’s private data being collected centrally.
It is a key technique for privacy-preserving AI.
A simple example
A phone keyboard’s next-word prediction can improve by learning from how millions of people type — but with federated learning, your actual messages never leave your phone. Only anonymous model updates are shared and combined.
Why it matters
Federated learning allows AI to improve from real-world data while protecting privacy, which is increasingly important as regulation and public concern around data grow. It offers a path to better models without centralising sensitive information.
Related terms
- Training Data — kept local in federated learning.
- Edge AI — related idea of on-device computation.
- Bias — a challenge federated learning must still manage.
Frequently asked questions
What is federated learning?
It trains a shared model across many devices or servers that keep their data local — only model updates, not the raw data, are shared — improving privacy.
Where is federated learning used?
It is used where data is sensitive or distributed, such as improving phone keyboards or healthcare models, without centralising private data.
