Skip to content
BENIBIX Learn

Federated Learning

Also known as Federated Machine Learning, Collaborative Learning (On-Device)

One model learns on many phones. Only the changes travel. The data stays put.

Think of it like

It is like a class where each child does homework at home. Only the answers go to school, not the books.

Example

Google tested it in 2017 on Gboard, its phone keyboard. Phones learned better word guesses from typing, and the typing stayed on each phone.

Why it matters

Some data is too private or too large to move. This lets a model learn from it anyway.

Definition

Federated learning trains one model across many devices without collecting their data. Each phone or hospital trains a copy of the model on its own data. It sends back only the changes to the model, not the data itself. A server mixes all the changes into one better model. Then it sends that model back out, and the loop repeats. The raw data never leaves home.
Only updates go up to the server. Each phone's own data stays where it is.

Common mistake

Thinking it makes the data fully private. Clever attacks can still learn things from the updates, so extra guards are added.

Key takeaways

  1. 1
    Train at home. Each device trains on its own data.
  2. 2
    Send changes only. Updates travel, data stays.
  3. 3
    Mix and repeat. A server blends them each round.

Learn it properly