Toward provably private learning from federated data

Federated Learning is a machine learning technique that trains models across decentralized private data while allowing devices to maintain control over their information. A new Federated Learning system uses Trusted Execution Environments to provide verifiable privacy guarantees, meaning third parties can confirm that data anonymization is happening as claimed. The system encrypts data on devices before uploading and requires pre-approved access policies that only release anonymized results. This approach improves training speed, accuracy, and device coverage compared to previous systems.