Models can learn from many devices without their data ever leaving home. Federated learning trains locally and shares only updates, keeping raw data private. It reshaped machine learning where privacy matters.
Train on device
Data stays put. Each phone or hospital trains on its own records. Nothing sensitive is sent.
Share only updates
Aggregation is central. A server combines model updates, not data. Privacy is preserved.
Privacy gains
The benefit is clear. Personal information never leaves the source. Exposure drops.
Communication cost
A trade-off appears. Frequent updates strain networks. Efficiency methods help.
Real deployments
Use is broad. Keyboards and health apps adopted it. Scale is large.
Remaining risks
Caution persists. Updates can still leak information. Added protections are studied.
The bottom line
Federated learning trains on-device and shares only model updates, keeping raw data private. It suits privacy-sensitive settings but strains communication. It is widely deployed with ongoing safeguards.
