Federated learning
Train one model across many hospitals without any of them sharing patient data — the model travels, the data stays.
In one line
Federated learning inverts the usual pipeline: instead of pooling sensitive data centrally, each site trains locally and shares only model updates, which a coordinator averages into a global model.
How it works
Rounds of: server sends current weights → each participating site trains a few steps on its own data → sites return weight updates → server aggregates (FedAvg) and repeats. Refinements handle real-world mess: secure aggregation so the server cannot inspect any single site's update, differential-privacy noise against reconstruction attacks, and techniques for non-IID data (every hospital's case mix is different — the central practical challenge).
Where it shows up in digital health
Multi-hospital imaging models where data-sharing agreements are impossible; rare-disease research across borders; consumer-device models trained on phones. It is a privacy-architecture, not a guarantee by itself — combine with DP and governance. For India-scale, multi-institution research under DPDP, this pattern is one to watch.