How it Works
Federated Learning
What is Federated Learning?
Federated learning is a collaborative approach to training machine learning models across multiple institutions. Instead of transferring data to a central location, each institution trains the model locally. Only model updates (not patient data) are shared and combined. This approach allows collaboration while maintaining institutional responsibility over data.
Learn more ➤Workflow
A typical collaborative experiment follows these steps and each step is structured and traceable
A research team defines a training plan.
Institutions review and approve participation.
Local training is performed within each participating institution.
Model updates are securely aggregated.
The improved model is redistributed.
Security in Practice
When a model is trained:
- The training happens locally inside the data provider’s secure environment.
- Only mathematical model updates are sent outside — never raw images or patient records.
- These updates are encrypted during transmission.
This means that even if someone intercepted the communication, they could not read or reconstruct patient data.
Original local data
never leaves the institution
Reconstruction attempt
Individual updates can reveal sensitive information.
without secure aggregation
Reconstruction attempt
Individual model updates are cryptographically protected before aggregation.
with secure aggregation