In 2017, Google introduced Federated Learning (FL), a machine learning technique that trains models across decentralized, private data. It powers features like next-word prediction and Smart Compose on Gboard, reply suggestions in Google Messages, and Smart Text Selection in Android. FL development is guided by four essential privacy principles: data minimization, data anonymization, transparency and control, and verifiability and auditability.
Years of research led to strong differential privacy (DP) guarantees using algorithms like MF-DP-FTRL and distributed DP with Secure Aggregation. In 2025, Google introduced an evolved FL definition centered on these principles. In “Toward provably private learning from federated data”, Google announces the next generation of its FL system, leveraging Trusted Execution Environments (TEEs) to provide fully verifiable and auditable data anonymization guarantees.
TEEs offer remotely attestable logic, confidentiality, and integrity. Gboard has already adopted the new system, benefiting from substantially faster compute times. The TEE-based FL system coordinates data upload, KMS and policy verification, workload execution, and fault-tolerant recovery. For more details, see the whitepaper.
Source: Google AI Blog · Summarized by HeadlinesBriefing