Paper 2024/476 Original: eprint.iacr.org/2024/476 ↗

OPSA: Efficient and Verifiable One-Pass Secure Aggregation with TEE for Federated Learning

Abstract

Federated learning enables collaborative model training while preserving data privacy by keeping data local. To protect user privacy during model aggregation, secure aggregation (SA) protocols are widely adopted to mask models. However, existing SA protocols require at least three round trips per aggregation and lack mechanisms to verify aggregation results. Verifiable SA addresses the verification gap but incurs high communication costs. TEE-based SA minimizes round trips but faces computational bottlenecks due to TEE's limited physical memory, especially when handling larger models or numerous clients. In this work, we introduce OPSA, an efficient and verifiable one-pass SA protocol based on TEE. By handling client dropouts via server-side TEE, OPSA enables the server to aggregate masked models in a single pass, significantly reducing round trips. To mitigate TEE's limitations, OPSA offloads tasks like model aggregation and mask elimination outside TEE, with only shared keys processed within TEE. Building on this design, we propose KhPRF-OPSA (single masking) and POT-OPSA (double masking) protocols, both incorporating novel cryptographic primitives. Furthermore, OPSA integrates commitment and signature mechanisms to ensure result verifiability with only $O(1)$ additional communication overhead per client. Compared to state-of-the-art schemes, OPSA achieves a 2$\sim$10$\times$ speedup in multi-round aggregation while guaranteeing result verification.