基于同态加密和秘密分享的纵向联邦LR协议研究

doi:10.12267/j.issn.2096-5931.2022.05.005

信息通信技术与政策 ›› 2022, Vol. 48 ›› Issue (5): 34-44.doi: 10.12267/j.issn.2096-5931.2022.05.005

基于同态加密和秘密分享的纵向联邦LR协议研究

Vertical federated logistic regression via homomorphic encryption and secret sharing

符芳诚¹^,², 刘舒², 程勇², 陶阳宇³

1.北京大学信息科学技术学院高可信软件技术重点实验室,北京 100871
2.腾讯TEG数据平台部,深圳 518054
3.腾讯TEG机器学习平台部,北京 100083

收稿日期:2022-03-10 出版日期:2022-05-15 发布日期:2022-05-26
通讯作者: 陶阳宇
作者简介:
符芳诚北京大学信息科学技术学院高可信软件技术重点实验室在读博士研究生,腾讯TEG数据平台部数据中心研究员,主要研究领域为机器学习、隐私计算、分布式计算等|刘舒腾讯TEG数据平台部数据中心高级研究员,主要研究领域为机器学习、隐私计算、分布式计算等|程勇腾讯TEG数据平台部数据中心专家研究员,主要研究领域为隐私计算、联邦学习、安全多方计算、分布式计算等

FU Fangcheng¹^,², LIU Shu², CHENG Yong², TAO Yangyu³

1. Department of Computer Science & Key Lab of High Confidence Software Technologies (MOE), Peking University, Beijing 100871, China
2. Data Platform, TEG, Tencent Inc., Shenzhen 518054, China
3. Machine Learning Platform, TEG, Tencent Inc., Beijing 100083, China

Received:2022-03-10 Online:2022-05-15 Published:2022-05-26
Contact: TAO Yangyu

摘要/Abstract

摘要：

提出了一种新颖的两方纵向联邦逻辑回归协议,并在半诚实安全模型下证明了该协议的安全性,包括模型训练流程和模型推理流程的安全性,且无需对非线性函数使用多项式近似计算,从而保证了联邦逻辑回归协议模型无损。

关键词: 纵向联邦学习, 逻辑回归, 同态加密, 秘密分享

Abstract:

This paper presents a novel vertical federated logistic regression algorithm with provable security guarantees of both model training and inference under the semi-honest security model. The proposed algorithm is privacy-preserving, lossless, and efficient. Firstly, by combining the homomorphic encryption and secret sharing mechanisms, data protection is provably ensured, including the protection of both features and labels. Secondly, the algorithm is lossless since it does not require any approximations for the non-linear functions.

Key words: vertical federated learning, logistic regression, homomorphic encryption, secret sharing

中图分类号:

TP309.2

符芳诚, 刘舒, 程勇, 陶阳宇. 基于同态加密和秘密分享的纵向联邦LR协议研究[J]. 信息通信技术与政策, 2022, 48(5): 34-44.

FU Fangcheng, LIU Shu, CHENG Yong, TAO Yangyu. Vertical federated logistic regression via homomorphic encryption and secret sharing[J]. Information and Communications Technology and Policy, 2022, 48(5): 34-44.

导出引用管理器 EndNote|Ris|BibTeX

链接本文:

http://ictp.caict.ac.cn/CN/10.12267/j.issn.2096-5931.2022.05.005

http://ictp.caict.ac.cn/CN/Y2022/V48/I5/34

图/表 10

图1 纵向联邦学习场景的数据划分示意图

图2 一种将同态密文转换为秘密分享变量的协议流程图

表1 同态密文转换为两个秘密分享变量的理想功能

输入	参与方A密钥sk_A
输入	参与方B输入密文$\left[\!\left[ \text{x} \right]\!\right]$和密钥pk_A
输出	参与方A输出x_A
输出	参与方B输出x_B,并满足明文空间里 x_A+x_B=x

图3 参与方A通过XAWA对标签进行预测的示意图

表2 纵向联邦逻辑回归算法协议中,为了保证数据安全, 各参与方不可获得的信息

	参与方A不应获得	参与方B不应获得
模型参数	W_A,W_B	W_A,W_B
前向计算	X_AW_A,X_BW_B,Z	X_AW_A,X_BW_B
反向计算	▽Z,▽W_A,▽W_B	▽W_A

图4 初始化流程图

图5 前向计算流程图

图6 反向计算流程图

表3 前向计算的理想功能

输入	参与方A输入特征X_A,模型${{\left[\!\left[ {{W}_{A}} \right]\!\right]}_{B}}$,和密钥sk_A,pk_A,pk_B
输入	参与方B输入特征X_B,模型U_B,${{\left[\!\left[ {{V}_{B}} \right]\!\right]}_{A}}$,和密钥sk_B,pk_A,pk_B
输出	参与方A无输出
输出	参与方B输出Z=X_AW_A+X_BW_B, $y ︿$ =Sigmoid(Z),其中W_B=U_B+V_B

表3 前向计算的理想功能

输入	参与方A输入特征X_A,模型${{\left[\!\left[ {{W}_{A}} \right]\!\right]}_{B}}$,和密钥sk_A,pk_A,pk_B
输入	参与方B输入特征X_B,模型U_B,${{\left[\!\left[ {{V}_{B}} \right]\!\right]}_{A}}$,和密钥sk_B,pk_A,pk_B
输出	参与方A无输出
输出	参与方B输出Z=X_AW_A+X_BW_B, $y ︿$ =Sigmoid(Z),其中W_B=U_B+V_B

表4 反向计算的理想功能

输入	参与方A输入特征X_A,模型${{\left[\!\left[ {{W}_{A}} \right]\!\right]}_{B}}$,和密钥sk_A,pk_A,pk_B
输入	参与方B输入特征X_B,标签y,预测值,模型U_B, ${{\left[\!\left[ {{V}_{B}} \right]\!\right]}_{A}}$,和密钥sk_B,pk_A,pk_B
输出	参与方A输出${{\left[\!\left[ \nabla {{W}_{A}} \right]\!\right]}_{B}}$
输出	参与方B输出$\nabla W_{B}$

参考文献 21

[1]	闫树, 袁博, 吕艾临. 隐私计算--推进数据“可用不可见”的关键技术[M]. 北京: 电子工业出版社出版, 2022.
[2]	中国信息通信研究院云计算与大数据研究所. 隐私计算白皮书(2021年)[R], 2021.
[3]	KONEN J, MCMAHAN B, RAMAGE D. Federated optimization: distributed optimization beyond the datacenter[J]. Mathematics, 2015.
[4]	YANG Q, LIU Y, CHEN T, et al. Federated machine learning: concept and applications[J]. ACM Transactions on Intelligent Systems and Technology, 2019, 10(2):1-19.
[5]	HARDY S, HENECKA W, IVEYLAW H, et al. Private federated learning on vertically partitioned data via entity resolution and additively homomorphic encryption[J], 2017. DOI: 10.48550/arXiv.1711.10677 doi: 10.48550/arXiv.1711.10677
[6]	YANG S, REN B, ZHOU X, et al. Parallel Distributed Logistic Regression for Vertical Federated Learning without Third-Party Coordinator[J]. arXiv:1911. 09824, 2019.
[7]	CHEN C, ZHOU J, WANG L, et al. When homomorphic encryption marries secret sharing: secure Large-Scale sparse logistic regression and applications in risk control[J]. Proceedings of the 27th ACM SIGKDD Conference on Knowledge Discovery & Data Mining, 2021(8):2652-2662.
[8]	RIVEST R L, ADLEMAN L M, DERTOUZOS M L. On data banks and privacy homomorphisms[J]. Foundations of Secure Compuation, 1978:169-180.
[9]	EVANS D, KOLESNIKOV V, ROSULEK M. A pragmatic introduction to secure multi-party computation[J]. Foundations & Trends© in Privacy & Security, 2018, 2(2-3):70-246.
[10]	PAILLIER P. Public-key cryptosystems based on composite degree residuosity classes[J]. Proc. EUROCRYPT’99, Czech Republic, May, 1999(4):223-238.
[11]	CHENG K, FAN T, JIN Y, et al. SecureBoost: a lossless federated learning framework[J]. Intelligent Systems, IEEE, 2021,(99):1-1.
[12]	FU F, SHAO Y, YU L, et al. VF 2 boost: very fast vertical federated gradient boosting for cross-enterprise learning[J]. Proceedings of the 2021 International Conference on Management of DataJune, 2021:563-576.
[13]	WU Y, CAI S, XIAO X, et al. Privacy Preserving Vertical Federated Learning for Tree-based Models[J], 2020. DOI: 10.14778/3407790.3407811. doi: 10.14778/3407790.3407811
[14]	ZHANG C, LI S, XUA J, et al. BatchCrypt: efficient homomorphic encryption for cross-silo federated learning[J]. In 2020 USENIX Annual Technical Conference, 2020(7):493-506
[15]	DEMMLER D, SCHNEIDER T, ZOHNER M. ABY-a framework for efficient mixed-protocol secure two-party computation[C]// Network & Distributed System Security Symposium, 2015.
[16]	MOHASSEL P, RINDAL P. ABY³: a mixed protocol framework for machine learning[J]. Proceedings of the 2018 ACM SIGSAC Conference on Computer and Communications SecurityOctober, 2018(10):35-52.
[17]	SHAMIR A. How to share a secret[J]. Communications of the ACM, 1979. DOI: 10.1145/359168.359176. doi: 10.1145/359168.359176
[18]	BEAVER D. Efficient multiparty protocols using circuit randomization[C]// Advances in Cryptology-CRYPTO’91, 11th Annual International Cryptology Conference, Santa Barbara, California, USA, Proceedings. Springer-Verlag, 1991.
[19]	PULLONEN P. Actively secure two-party computation: efficient beaver triple generation[Z], 2013.
[20]	GOLDREICH O. The foundations of cryptography - volume 2, basic applications[M], 2004.
[21]	LINDELL Y. How to simulate it - a tutorial on the simulation proof technique[J]. Springer International Publishing, 2017:277-346.

基于同态加密和秘密分享的纵向联邦LR协议研究

Vertical federated logistic regression via homomorphic encryption and secret sharing

RichHTML

PDF

可视化

摘要/Abstract

引用本文

使用本文

图/表 10

参考文献 21

相关文章 3

编辑推荐

Metrics

本文评价

[1]	柴迪. 阈值同态加密在隐私计算中的应用[J]. 信息通信技术与政策, 2021, 47(7): 82-86.
[2]	夏家骏, 鲁颖, 张子扬, 张钰婷, 张佳辰. 基于秘密共享与同态加密的纵向联邦学习方案研究[J]. 信息通信技术与政策, 2021, 47(6): 19-26.
[3]	冯浩铭, 　陈杭, 　刘彦平. 一种金融级安全的区块链资产交易系统[J]. 信息通信技术与政策, 2020, 46(1): 64-68.