Skip to content
AI & analytics

Federated learning

A training approach where devices improve a shared model locally and send only parameter updates to a server, so raw data never leaves the device.

In federated learning each participating device trains on its own data and uploads model updates, which a coordinator aggregates into a new global model that is sent back out. The raw readings, which may be audio, health signals or usage logs, stay local. This addresses data sovereignty constraints and reduces upstream bandwidth compared with shipping every sample to the cloud.

The approach suits fleets of similar devices, such as wearables or keyboards learning user patterns. The difficulties are practical: devices have uneven data, intermittent connectivity and little spare compute, and the updates themselves can leak information unless they are protected. It complements rather than replaces centralized model training and still needs an OTA update path to deliver the aggregated model.

Part of the IoT Glossary. Spotted an error or a missing term? Tell us.