Glossary
Federated learning
A way of training a machine-learning model across several organizations without moving their raw data to a central place.
In conventional machine learning, all training data is pooled in one location. In federated learning, the model travels instead: each participating site trains on its own data locally and shares only model updates (weights or gradients), which are aggregated into a shared model. Raw records never leave the site.
The approach suits healthcare, where data is sensitive, governed by different institutions, and often cannot be shared. It lets hospitals or clinics contribute to a common model while keeping custody of their own data. It is not a complete privacy solution on its own — updates can leak information unless additional safeguards are used — but it removes the single biggest barrier: centralizing patient data.
Why it matters to our work
Scalable, reproducible analytics across sites is part of how we think about extending our neuroanalytics framework, with federated approaches where they are appropriate and feasible.