Federated Learning
Train a model across dozens of devices without ever centralizing the raw data.
Think of it like
Like a book club where everyone reads at home and reports their thoughts, instead of mailing every book to one person first.
Example
A smartphone keyboard learns from millions of users typing text without sending anyone's messages to a central server—only the model updates.
How it actually works
Each device trains locally on its own data, then sends only the updated model weights back to a server. The server merges those updates and sends the improved model back out. No raw data leaves the device, which is huge for privacy. Catch: communication cost is high, and heterogeneous data (different people's typing styles) can confuse the learning.
For product teams
Let users keep their data private while still getting the benefit of collective learning.
For engineers
Implement parameter averaging across devices, manage stragglers and dropout, and optimize communication bandwidth.
Related
- Privacy — Complements privacy goals.
- Differential Privacy — Often combined with privacy techniques.
Read anything AI without the jargon
Look up any term in plain English, or save terms as you read with the free Chrome extension.
Open DecoderAdd to Chrome