高效可扩展的多密钥同态秘密分享协议OA
Efficient and Scalable Multi-Key Homomorphic Secret Sharing Protocol
同态秘密分享是一种支持参与者在计算服务器间共享私有数据并联合完成公共函数计算的数据隐私保护技术,其优势在于无需依赖全同态加密.然而,现有可处理多源数据的同态秘密分享方案存在显著局限性:部分方案仅能计算次数受参与者数量制约的多项式,而另一些方案则依赖于复杂的多密钥同态加密技术,导致效率低下且难以扩展.本文基于 Xu 等人提出的多密钥同态秘密分享形式化概念,结合容错学习问题,提出了一种高效且可扩展的多密钥同态秘密分享方案,能够在服务器的辅助下对来自多个独立数据源的私有数据进行受限乘法直线程序的安全计算.该方案不依赖传统的多密钥全同态加密,通过构造一种加密方案 ERegev,将密文矩阵从 Zd×2lq 降为了 Zd×2q,使得加密效率显著提升;在客户端进行预计算以减少通信量.能够在服务器的辅助下对来自多个独立数据源的私有数据进行受限乘法直线程序的安全计算.在安全性方面,基于 LWE 假设,本文给出了在半诚实敌手模型下严格的安全性证明.在性能方面,实验结果表明,与 Xu 等人的方案相比,本文方案在通信开销上降低了超过 90%,服务器端算法执行效率平均提升 4 倍以上,客户端算法的执行效率提升 2 个数量级,显著优化了资源消耗与计算效率.本文工作为多源数据隐私计算场景提供了一种更高效、更实用的解决方案.
Homomorphic secret sharing is a data privacy protection technology that enables partic-ipants to share private data among computing servers and collaboratively compute public functions without relying on fully homomorphic encryption.However,existing homomorphic secret sharing schemes capable of handling multi-source data face significant limitations:some can only evaluate polynomials whose degrees are constrained by the number of participants,while others rely on com-plex multi-key homomorphic encryption technologies,leading to low efficiency and poor scalability.To address these issues,this paper proposes an efficient and scalable multi-key homomorphic secret sharing scheme based on the formal concept of multi-key homomorphic secret sharing introduced by Xu et al.,combined with the learning with errors problem,which can compute securely restricted multiplication straight-line programs on private data from multiple independent sources with server assistance.This scheme abandons the traditional reliance on multi-key fully homomorphic encryption and employs homomorphic linear combination techniques to securely compute restricted multiplication straight-line programs on private data from multiple independent sources with server assistance.In terms of security,this paper provides rigorous security proofs to ensure the scheme's security under standard security assumptions.This scheme does not rely on the traditional multi-key fully homomor-phic encryption.By constructing an encryption scheme ERegev,the ciphertext matrix is reduced from Zd×2lq to Zd×2q,significantly improving the encryption efficiency;pre-computation is carried out at the client side to reduce the communication volume.In terms of security,based on the LWE assumption,this paper provides a strict security proof under the semi-honest adversary model.Performance-wise,experimental results demonstrate that compared to Xu et al.'s scheme,our scheme reduces communica-tion overhead by over 90%and improves the execution efficiency of Server-side algorithm by an average of more than 4 times,and the execution of the client-side algorithm has achieved an improvement in efficiency by two orders of magnitude,significantly optimizing resource consumption and computa-tional efficiency.This work offers a more efficient and practical solution for multi-source data privacy computing scenarios.
张明广;黄豪杰;刘亮;张书卿;高莹
北京航空航天大学 网络空间安全学院,北京 100191北京航空航天大学 网络空间安全学院,北京 100191北京航空航天大学 网络空间安全学院,北京 100191北京航空航天大学 网络空间安全学院,北京 100191北京航空航天大学 网络空间安全学院,北京 100191
信息技术与安全科学
隐私保护同态秘密分享容错学习问题同态线性组合受限乘法直线程序
privacy preservationhomomorphic secret sharinglearning with errorshomomorphic linear combinationrestricted multiplicative linear programs
《密码学报(中英文)》 2026 (4)
689-708,20
国家自然科学基金专项项目(12441101)Special Program of National Natural Science Foundation of China(12441101)
评论