Federated Learning
Last Updated: July 29, 2026 | By Mihail Sebastian | AI Dictionary
A machine learning approach that trains a shared model across many devices while raw data stays on each device; only model updates travel to a server.
What is Federated Learning?
Federated learning trains a shared machine learning model across many devices or organizations without collecting their data in one place. Each participant trains locally on its own data and sends only model updates to a central server; the raw data never leaves the device.
That property makes it a privacy technique as much as a training technique. It enables model training on data that is too sensitive, too regulated, or too large to centralize.
How Federated Learning Works
Training runs in rounds. The server sends the current model to participating devices, each device trains it briefly on local data, and the devices return their updated weights. The server averages those updates into a new global model, and the cycle repeats until the model converges.
Sharing updates instead of data shrinks the privacy risk but does not erase it: researchers have shown that model updates can leak information about the data behind them. Serious deployments pair federated learning with safeguards such as differential privacy or secure aggregation.
Types of Federated Learning
- Horizontal federated learning: Participants hold the same kinds of features about different people, like several hospitals recording the same patient attributes.
- Vertical federated learning: Participants share many of the same people but hold different features about them, like a bank and a retailer with overlapping customers.
- Federated transfer learning: Participants overlap in neither samples nor features, and transfer learning bridges the gap between their models.
Applications of Federated Learning
- Healthcare: Hospitals jointly train diagnostic models without exchanging patient records, which would be hard to square with privacy law.
- Finance: Banks build shared fraud detection models while each keeps client data in-house, supporting compliance with confidentiality rules.
- Smartphones: Vendors improve on-device features like predictive text by learning from millions of phones without reading anyone’s messages.
Example of Federated Learning
Google’s Gboard keyboard is the canonical deployment. Each phone trains the next-word prediction model on what its owner actually types, and does so while the phone is idle, charging, and on Wi-Fi, so the user never notices.
The phone sends only the resulting model update to Google’s servers, where it is averaged with updates from many other phones. Every keyboard gets better at prediction; no one’s typing history leaves their device.
Related AI terms: Edge Computing · Model Training · Transfer Learning · Compliance
Did you like the Federated Learning gist?
Learn about 250+ need-to-know artificial intelligence terms in the AI Dictionary.
Mihail Sebastian — Writes about AI governance, regulation, and the technology behind them. Placeholder bio — replace with a real credential line. About