Federated learning

Federated Learning (FL) trains a shared AI model across numerous decentralized devices, keeping all raw training data local to ensure privacy and data sovereignty.

FL is a distributed machine learning paradigm that decouples model training from centralized data collection. The process is iterative: a central server sends the current global model to a subset of clients (e.g., smartphones, hospital servers). Each client trains the model locally on its private data, then sends only the model updates (weights or gradients) back to the server, never the raw data. The server aggregates these updates, typically using the Federated Averaging (FedAvg) algorithm, to create an improved global model. This approach directly addresses critical regulatory concerns like GDPR and HIPAA, enabling collaborative AI development in sensitive sectors like healthcare and powering on-device features, such as Google's Gboard predictive text.

https://ai.google/research/federated-learning/