保护私有信息的回归关系推断  

Privacy-preserving Inference of Regression Relationship

作　　者：姚亦飞[1,2] 黄刘生[1,2] 徐维江[1,2] 荆巍巍[1,2] 

机构地区：[1]中国科学技术大学计算机科学技术系,安徽合肥230027 [2]安徽省计算与通讯软件重点实验室,安徽合肥230027

出　　处：《小型微型计算机系统》2009年第11期2128-2132,共5页Journal of Chinese Computer Systems

基　　金：国家自然科学基金项目(6077303260703071)资助;高校博士点基金项目(20060358014)资助;江苏省自然科学基金项目(BK2007060)资助;安徽省自然科学基金项目(070412043)资助

摘　　要：回归关系推断是统计分析领域中的重要基础内容,其运算形式较为复杂,导致了安全解决方案的效率低下.为了解决安全双方模型下多元回归的统计推断问题,提出极大似然估计、残差平方和与总平方和的安全协议.并依据运算本身的特点,经过对共享矩阵连乘操作规律的数学分析,提出简化运算的方法,使得交互次数大幅减少.同时对其可应用领域进行了展望.Inference of regression relationship is a fundamental content in statistical analysis. Its operational form is somehow too complicated to be computed efficiently. In order to solve multiple linear regression under secure two-party model , the secure protocols for maximum likelihood, sum of squared residuals and regression sum of square are proposed. According to their specialties of calculation, the simplification discipline of matrix continued product is summarized and the number of interactive time is reduced. At last, the applications for privacy-preserving predication detection are also discussed.

关 键 词：安全双方计算 保护私有信息的统计分析 回归关系 矩阵求逆 

分 类 号：TP309[自动化与计算机技术—计算机系统结构]