基于暹罗网络的云计算隐私保护算法  

Cloud Computing Privacy Protection Algorithm Based on Siamese Network

作　　者：朱利华[1] 朱玲玲[2] ZHU Li-hua;ZHU Ling-ling(School of Software and Big Data, Changzhou College of Information Technology, Changzhou Jiangsu 213164, China;School of Information Science and Technology, Nantong University, Nantong Jiangsu 226200, China)

机构地区：[1]常州信息职业技术学院软件与大数据学院,江苏常州213164 [2]南通大学信息科学技术学院,江苏南通226200

出　　处：《西南师范大学学报（自然科学版）》2021年第7期84-89,共6页Journal of Southwest China Normal University(Natural Science Edition)

基　　金：国家自然科学基金项目(61673384);全国高等院校计算机基础教育研究会重点课题(GZYZD2018014).

摘　　要：针对云计算难以在准确性和隐私性之间取得可接受的折衷问题,提出一种基于暹罗网络的云计算隐私保护移动分析算法.该算法通过暹罗网络对特征提取模块进行微调来选择适用于主要任务,但不适用于其他辅助任务的专有特征,采用主成分分析(Principle Component Analysis,PCA)法对专有特征进行降维,增加其私密性,降低通信开销,通过在特征向量里嵌入多维噪声来进一步提高隐私性.最后,基于对任意敏感变量统计分析的隐私评估方法评估隐私和验证专有特征提取的质量.实验结果表明:本文提出的框架可以在实用性、隐私性和准确性之间实现理想的平衡.Aiming at the problem that cloud computing is difficult to achieve an acceptable compromise between accuracy and privacy,a mobile analysis algorithm for cloud computing privacy protection based on Siam Network has been proposed.The algorithm uses the Siamese network to fine-tune the feature extraction module to select proprietary features that are suitable for the main task but not suitable for other auxiliary tasks.Principal component analysis(PCA)is used to reduce the dimension of the special features to increase its privacy and reduce the communication overhead.The privacy is further improved by embedding multi-dimensional noise into the feature vector.Finally,a privacy evaluation method based on statistical analysis of any sensitive variable is used to evaluate privacy and verify the quality of proprietary feature extraction.Experimental results show that the proposed framework can achieve an ideal balance between practicality,privacy and accuracy.

关 键 词：云计算 隐私保护 暹罗网络结构 深度学习 

分 类 号：TP393[自动化与计算机技术—计算机应用技术]