Probability matrix decomposition of personalized differential privacy protection

Time: 2019-12-09 15:41:00

Topic：Probability matrix decomposition of personalized differential privacy protection

Speaker：Shun Zhang      Anhui University

Time：2019-12-09 17:30--18：30

Location：Room 306 in School of Mathematical Sciences

Introduction：

     In recent years, differential privacy has become a research hotspot in the intersection of machine learning and information security. It can protect individual privacy information of users by adding controllable noise without changing the overall pattern characteristics of data, so as to meet the privacy protection and data utility requirements. This report first introduces a recommendation scheme based on general difference privacy protection (dp-pmf) by combining the probability matrix decomposition recommendation algorithm. Furthermore, based on the differentiated needs of user groups for privacy protection, an improved sampling mechanism was developed, scoring data was expressed by matrix, and a pdp-pmf recommendation scheme based on personalized differential privacy protection was proposed. From the theoretical point of view, the scheme realizes project-level differential privacy protection and meets users' personalized privacy protection needs. In addition, we conducted a series of comparative experiments in the allocation of multiple user levels and privacy budgets, as well as the optimization of sampling thresholds, which verified the superiority of pdp-pmf scheme in recommendation accuracy.