Federated Learning

Federated learning trains a shared model across many devices or organizations without centralizing the raw data. Each participant trains locally on its own data, sends only model updates (gradients or weight deltas) to a coordinating server, and the server averages them — federated averaging — into an improved global model that is redistributed for the next round. Google pioneered it for Gboard's next-word prediction; hospitals and banks use the cross-silo variant to learn from data that regulation forbids pooling. It is not automatically private: gradients can leak training examples, so serious deployments layer on differential privacy and secure aggregation. Engineering challenges are real — devices drop out, data is wildly non-IID across participants, and communication is the bottleneck. For SaaS builders it matters as a compliance-friendly answer to "train on customer data without ever holding it," particularly in healthcare, finance, and on-device products.