Keyboards that learn without watching
One of the clearest production use cases for federated learning is next-word prediction on mobile keyboards. A keyboard app benefits enormously from learning the patterns of how people actually type, but the raw text people type is some of the most sensitive data on a phone. Federated learning lets the underlying prediction model improve using patterns learned locally on-device, with only model updates, not typed text, ever leaving the phone.
This use case also connects naturally to edge AI, since the training itself happens on-device using local compute rather than in the cloud. The two ideas are complementary: edge AI is about running or training models on local hardware, while federated learning is specifically about how many such local training sites can cooperate to build one shared model without pooling their data.