What is a core principle of federated learning?

A core principle of federated learning is that models are trained without transferring raw data. In this framework, the learning process takes place locally on individual devices, such as smartphones or IoT devices, which hold their own user data. Instead of sending raw data to a central server for processing, each device computes updates to the model based on its local data. These updates are then sent to the central server, where they are aggregated to improve the global model. This approach enhances privacy and security since personal data remains on the devices and is not exposed to central storage.

The other options do not align with the principles of federated learning. Training on a central server contradicts the decentralized nature of the method. Sharing raw data among devices would compromise user privacy, which federated learning aims to protect. Lastly, the requirement for continuous connectivity is not a necessity for federated learning; devices can often work offline and then sync when they connect to a network.