What Is Federated Learning? Training a Model Without Centralizing the Data
A training approach where a model trains across many separate devices or organizations, without their raw data ever having to leave its original location.
Training where the data already lives, not in one central place
Federated learning trains a shared model across many separate devices or organizations by having each one train locally on its own data, then sending only the resulting model updates — not the raw data itself — back to be combined centrally into an improved shared model.
Why keeping data decentralized matters
In many real settings — hospitals with patient records, phones with personal messages — sending raw data to a central location for training raises real privacy and legal concerns. Federated learning lets a shared model still benefit from that distributed data's patterns, without the raw, sensitive data ever having to leave its original, local location.
The real tradeoffs involved
Coordinating training across many separate devices or organizations, each potentially with different amounts and kinds of local data, is meaningfully more complex than standard centralized training, and it typically produces somewhat less consistent results than training would on fully centralized data — a genuine tradeoff for the privacy benefit gained.
Frequently Asked Questions
What is federated learning?
A training approach where a shared model trains across many separate devices or organizations, each training locally on its own data and sending back only model updates — not raw data — to be combined centrally.
Why is federated learning used instead of centralizing all the data?
It avoids sending sensitive raw data — like hospital records or personal messages — to a central location, letting a shared model still benefit from that distributed data's patterns without the privacy and legal concerns of centralizing it.