基于CKKS同态加密的纵向联邦逻辑回归训练方案  

A Vertical Federated Logistic Regression Training Scheme Based on CKKS Homomorphic Encryption

作　　者：汤殿华 曹云飞 黄云帆[1,3] 李枫 TANG Dianhua;CAO Yunfei;HUANG Yunfan;LI Feng(National Key Laboratory of Security Communication,Chengdu Sichuan 610041,China;School of Computer Science and Engineering,University of Electronic Science and Technology of China,Chengdu Sichuan 611731,China;No.30 Institute of CETC,Chengdu Sichuan 610041,China)

机构地区：[1]保密通信全国重点实验室,四川成都610041 [2]电子科技大学计算机科学与工程学院,四川成都611731 [3]中国电子科技集团公司第三十研究所,四川成都610041

出　　处：《通信技术》2024年第11期1173-1180,共8页Communications Technology

基　　金：国家重点研发计划(2023YFB3106200)。

摘　　要：当前,数据泄露、模型窃取事件频发,机器学习隐私保护成为安全领域的重要议题。联邦学习能够在原始数据不出本地的情况下,实现分布式训练和推理,是实现机器学习隐私保护的关键技术之一。针对基于单同态加密纵向联邦学习方案存在的训练耗时较高、通信开销大等问题,采用CKKS同态加密方案,并利用打包编码的批处理优势,设计了基于CKKS同态加密的联邦纵向逻辑回归训练方案,经实验表明,相比于FATE的基于Paillier的纵向逻辑回归训练方案,所提方案效率提升约14.12倍。Currently,with the frequent occurrence of data leakage and model theft,privacy protection in machine learning becomes an important issue in the security field.Federated learning can achieve distributed training and reasoning without losing original data,and it is one of the key technologies to solve privacy protection issues in machine learning.To address the problems of long training time and high communication overhead of vertical federated learning scheme based on single homomorphic encryption,this paper adopts the CKKS homomorphic encryption scheme and takes advantage of the batch processing of packaged encoding to design a federated vertical logistic regression training scheme based on CKKS homomorphic encryption.Experimental results indicate that compared with FATE’s Paillier-based vertical logistic regression training scheme,the efficiency of the proposed scheme is improved by about 14.12 times.

关 键 词：联邦学习 同态加密 隐私保护 批处理 

分 类 号：TP309.2[自动化与计算机技术—计算机系统结构]