Back to Glossary
Cloud
What is Federated Learning?
A machine learning approach where the model is trained across multiple devices without raw data leaving each device, preserving data privacy.
Federated learning allows AI to learn from data without collecting it centrally.
How It Works
- A model is sent to each user's device
- The model trains on local data
- Only model updates (gradients) are sent back — not the raw data
- Updates are aggregated to improve the global model
- The improved model is sent back to devices
Privacy Benefits
- Raw data never leaves the device
- Reduces central data collection
- Can be combined with differential privacy for stronger guarantees
Limitations
- Model updates can still leak information about the training data
- Gradient inversion attacks can reconstruct training data from updates
- Requires trusting the aggregation server
Where It's Used
- Google Gboard (keyboard prediction without uploading typed text)
- Apple's on-device ML improvements
- Health research across hospitals without sharing patient records
Related Terms
Differential Privacy
A mathematical framework for sharing aggregate information about a dataset while provably protecting the privacy of individual entries.
Secure Multi-Party Computation
A cryptographic technique that allows multiple parties to jointly compute a function over their combined data without revealing their individual inputs to each other.
Have more questions?
Use our guided flow to get the right next privacy step for Federated Learning.
Open Guided Flow