What is Federated Learning

Definition

Federated learning is a machine-learning approach that trains a shared model across many decentralised devices or servers holding local data, sending only model updates rather than raw data to a central coordinator, so models improve without the data ever leaving its source.

« Back to Glossary Index

Trains models without moving sensitive raw data off devices
Preserves privacy by sharing only model updates, not data
Leverages data spread across many devices or organisations
Reduces data-transfer cost and central storage of sensitive data

Real World Example

A phone keyboard improves its next-word prediction through federated learning, training on each user's typing locally and sending only anonymised model updates, so the model gets smarter without users' text ever leaving their phones.

FAQs

How does federated learning protect privacy?

Raw data stays on each device; only model updates are shared and aggregated, so personal data is never centrally collected.

What is the central server's role?

It coordinates training by distributing the model and aggregating the updates received from participating devices.

What challenges does federated learning face?

Non-uniform device data, communication overhead, and ensuring updates themselves do not leak information are key challenges.