A general framework for high-dimension data secure aggregation with resilience to dropouts

Data secure aggregation (DSA) protocols play an important role in many applications with privacy preservation, e.g., medical data analysis, federated learning model aggregation, etc. In such protocols, the computation and communication complexity of clients and the aggregator heavily depend on two parameters, including the number of involved clients and the dimension of secret data. Besides, resilience to client dropouts is an crucial requirement in many applications. In this paper, we focus on the issue of high-dimension DSA (HDDSA) with resilience to dropouts. Based on a critical non-interactive masking method using lightweight computations over polynomials, we propose a DSA reduction framework to transform high-dimension DSA problem to secure aggregation over scalars. We also construct two efficient HDDSA protocol instantiations based on multiparty homomorphic encryption (MPHE) cryptosystems. The first one (HDDSA1) gives a 2-round DSA protocol based on a threshold Paillier’s cryptosystem which requires a trusted setup. The second one (HDDSA2) gives a 3-round DSA protocol based on a multiparty Brakerski–Fan–Vercauteren (MPBFV) cryptosystem, which by contrast does not need trusted setup. Both protocols are resilient to dropouts by design and do not introduce extra recovery overheads. In addition, both protocols are secure against semi-honest adversary and collusion adversary with up to $\min (t-1,n-2)$ clients, given $n$ clients involved in the protocols, $t$ is a threshold parameter of underlying subprotocol. In terms of efficiency, the computation and communication complexity at client side are both $O\left(\ell \right)$, where $\ell$ is the dimension, which is independent of the number of clients. Empirical experiments are also conducted to show the practical efficiency superiority of our framework and proposed protocols.