Federated learning

Federated learning trains a shared model across multiple decentralized devices or data silos without moving the underlying data, useful when data residency or privacy constraints prevent centralization.

What is federated learning?

Federated learning is a distributed ML training pattern where multiple parties collaboratively train a shared model without sharing the underlying training data. Each party trains locally on their data and sends only model updates (gradients or weight deltas) to a central aggregator, which combines them into the next round's shared model. Originally developed for mobile keyboard prediction (Google, 2017), now applied to healthcare, finance, and cross-jurisdictional AI use cases.

Privacy properties

FL alone is not a privacy guarantee — model updates can leak training data through gradient inversion attacks. Production FL deployments layer additional protections: differential privacy on the aggregation, secure aggregation cryptography, and bounded participation per round. Without these, FL is mostly a data-residency tool, not a privacy tool. With them, it is genuinely useful for combining data across regulatory boundaries (e.g. multi-hospital research consortia).

When buyers see FL

Healthcare AI vendors offering "train on your data without it leaving your environment" usually mean FL. Financial-services compliance teams sometimes require FL for cross-jurisdictional model training. Ask: what privacy enhancements are layered on top of vanilla FL, what is the aggregation protocol, what happens to model updates after aggregation, and have the privacy guarantees been audited externally.