Page 339 - 《软件学报》2026年第6期

软件学报 ISSN 1000-9825, CODEN RUXUEW E-mail: jos@iscas.ac.cn
2026,37(6):2658−2670 [doi: 10.13328/j.cnki.jos.007475] [CSTR: 32375.14.jos.007475] http://www.jos.org.cn
©中国科学院软件研究所版权所有. Tel: +86-10-62562563

*
基于秘密分享的高效隐私保护卷积神经网络预测

白 浩 1,2 , 何 琨 1,2 , 陈 晶 1,2 , 赵陈斌 1,2 , 杜瑞颖 1,2

(空天信息安全与可信计算教育部重点实验室 (武汉大学), 湖北 武汉 430040)
1
2
(武汉大学 国家网络安全学院, 湖北 武汉 430040)
通信作者: 何琨, E-mail: hekun@whu.edu.cn

摘 要: 针对隐私保护卷积神经网络预测, 先前的研究采用同态加密、安全多方计算等方法来保护客户端隐
私敏感数据. 然而, 这些方法通常面临预测时间开销过大的问题. 为了解决此问题, 提出一个高效的隐私保护
卷积神经网络预测方案. 该方案根据卷积神经网络中线性层和非线性层不同计算特点, 设计矩阵分解计算协
议和参数化二次多项式近似 ReLU 激活函数方法, 从而实现了线性层和非线性层高效安全计算, 并缓解了近似

处理而导致的预测准确率损失. 在线性层和非线性层中的计算都可以通过轻量级密码原语秘密分享来完成.
理论分析和实验结果表明, 在保证安全性前提下, 所提方案将预测速度提高了 2–15 倍, 同时预测准确率损失
仅约为 2%.
关键词: 隐私保护预测; 卷积神经网络; 秘密分享; 非线性近似; 深度学习
中图法分类号: TP309

中文引用格式: 白浩, 何琨, 陈晶, 赵陈斌, 杜瑞颖. 基于秘密分享的高效隐私保护卷积神经网络预测. 软件学报, 2026, 37(6):
2658–2670. http://www.jos.org.cn/1000-9825/7475.htm
英文引用格式: Bai H, He K, Chen J, Zhao CB, Du RY. Efficient Privacy-preserving Inference Based on Secret Sharing for
Convolutional Neural Network. Ruan Jian Xue Bao/Journal of Software, 2026, 37(6): 2658–2670 (in Chinese). http://www.jos.org.cn/
1000-9825/7475.htm

Efficient Privacy-preserving Inference Based on Secret Sharing for Convolutional Neural Network
1,2
1,2
1,2
1,2
BAI Hao , HE Kun , CHEN Jing , ZHAO Chen-Bin , DU Rui-Ying 1,2
1
(Key Laboratory of Aerospace Information Security and Trusted Computing (Wuhan University), Ministry of Education, Wuhan 430040,
China)
2
(School of Cyber Science and Engineering, Wuhan University, Wuhan 430040, China)
Abstract: In privacy-preserving inference using convolutional neural network (CNN) models, previous research has employed methods
such as homomorphic encryption and secure multi-party computation to protect client data privacy. However, these methods typically
suffer from excessive prediction time overhead. To address this issue, an efficient privacy-preserving CNN prediction scheme is proposed.
This scheme exploits the different computational characteristics of the linear and non-linear layers in CNNs and designs a matrix
decomposition computation protocol and a parameterized quadratic polynomial approximation for the ReLU activation function. This
enables efficient and secure computation of both the linear and non-linear layers, while mitigating the prediction accuracy loss caused by
the approximations. The computations in both the linear and non-linear layers can be performed using lightweight cryptographic primitives,
such as secret sharing. Theoretical analysis and experimental results show that, while ensuring security, the proposed scheme improves
prediction speed by a factor of 2 to 15, with only about a 2% loss in prediction accuracy.
Key words: privacy-preserving inference; convolutional neural network (CNN); secret sharing; non-linear approximation; deep learning

* 基金项目: 国家重点研发计划 (2022YFB3102100); 国家自然科学基金 (62172303, 62302343, 62076187); 湖北省重点研发计划 (2021BAA190,
2022BAA039); 山东省重点研发计划 (2022CXPT055)
收稿时间: 2024-11-11; 修改时间: 2025-03-17; 采用时间: 2025-05-26; jos 在线出版时间: 2025-09-28
CNKI 网络首发时间: 2025-09-30