Toward verifiably private learning from federated data
[Kryptografia i bezpieczeństwo] Autorzy: Katharine Daly, Yu Xiao, Zachary Garrett i 9 więcej
Federated Learning (FL) allows devices with private data to collaborate in training a shared model. We present a next-generation FL system based on Trusted Execution Environments (TEEs) that addresses operational challenges associated with earlier systems and provides externally verifiable central Differential Privacy (DP) guarantees for the first time while offering a better privacy-utility tradeoff. In our system, devices upload data encrypted with keys managed by a TEE-hosted Key Management Service (KMS). The uploaded data is cryptographically tied to a policy limiting the set of Python programs that may later process the data in server-side TEEs. External parties may inspect public transparency logs to observe the set of workloads allowed by these policies. Our experimental results show that the new system improves device coverage and favorably shifts privacy-utility curves by enabl…
- Kategoria arxiv
- cs.CR
- Kategoria
- Kryptografia i bezpieczeństwo