一种基于随机掩码的低通信量Logistic回归外包训练方案  

An Outsourcing Training Scheme of Low Communication Logistic Regression Based on Random Mask

作　　者：黄晓文 王政杰 崔硕硕 张宇浩 邓国强[1] HUANG Xiaowen;WANG Zhengjie;CUI Shuoshuo;ZHANG Yuhao;DENG Guoqiang(School of Mathematics&Computing Science,Guilin University of Electronic Technology,Guilin,Guangxi Zhuang Autonomous Region,541004 China)

机构地区：[1]桂林电子科技大学数学与计算科学学院,广西桂林541004

出　　处：《科技资讯》2021年第34期5-9,共5页Science & Technology Information

基　　金：广西科技基地和人才专项(桂科AD18281024);广西区级大学生创新训练计划项目(项目编号:S202110595217)资助。

摘　　要：Logistic回归是一种典型的机器学习模型,因其在疾病诊断、金融预测等许多应用表现优越而受到广泛关注。Logistic回归模型的建立不仅依赖于算法,更依赖于大量有效的训练数据。尽管构建高精度模型并提供预测服务有诸多优点,但用户的敏感信息数据造成隐私问题。因此,该文提出一个新的Logistic回归外包训练方案。在该方案中,用户会预先对私有数据进行处理,并添加随机掩码的数据矩阵上传给聚合器,聚合器将聚合得到的全局训练矩阵上传给云服务器进行训练。该方案在满足数据隐私的安全性需求下具有较高的计算效率和较低的通信开销。Logistic regression is a typical machine learning model,which has attracted extensive attention due to its superior performance in many applications such as disease diagnosis and financial prediction.The establishment of Logistic regression model depends not only on the algorithm,but also on a large number of effective training data.Although building high-precision models and providing prediction services have many advantages,users'sensitive information and data cause privacy problems.Therefore,a new outsourcing privacy protection Logistic training framework is proposed in this paper.In this scheme,the user will process the private data in advance,and upload the data matrix with random mask to the aggregator,and the aggregator will upload the aggregated global training ma‐trix to the cloud server for training.This scheme has high computational efficiency and low communication over‐head while meeting the security requirements of data privacy.

关 键 词：LOGISTIC 回归 隐私保护 随机掩码 低通信量 

分 类 号：TP309[自动化与计算机技术—计算机系统结构]