Federated learning is a type of machine learning where data is distributed among different devices, instead of being centralized in a single server. Devices can be trained locally on their own data, and then share their model updates with each other, without sharing the underlying data.