Collaborative Visual Analytics Method Based On Secure Multi-party T-SNE

Nowadays,there is a growing problem of”data silos”,where data is often distributed among many different data owners.All data own-ers hope to cooperate with each other to integrate data resources for analysis on the premise of protecting their own data privacy,so as to obtain more valuable information.In the field of visualization,a collaborative visual analytics method is proposed to break down this data barrier.However,when multiple privacy-sensitive data parties are combined for collabora-tive visual analytics,the traditional collaborative visual analytics method may expose data privacy.In order to solve the above problems,this thesis proposes a cooperative visual analytics method based on secure multi-party t-SNE.In this thesis,the main research contents include:(1)This thesis proposes a secure multi-party t-SNE projection method.This method provides a global analysis environment for collaborative vi-sual analysis while protecting privacy.This thesis use a secure multi-party computing framework based on homomorphic encryption and Braille tech-nology commonly used in secure multi-party computing to reconstruct the t-SNE algorithm.However,this framework can not guarantee the security of data privacy and complete the projection calculation at the same time.Therefore,this thesis improves the noise addition scheme of braille technol-ogy in the framework,and proposes a consistent Braille technology,which can protect data privacy while calculating accurate t-SNE projection.The time complexity of secure multi-party t-SNE is the same as that of standard t-SNE,which is O(n~2m),where n is the number of data points and m is the dimension of data points.(2)This thesis designs and implements a secure multiparty projection system(SMAP)to organize secure multiparty projection tasks among mul-tiple participants.SMAP system organizes secure multi-party projection tasks between multiple participants and supports the exploration and anal-ysis of joint projections while protecting data privacy.The system imple-ments two different levels of visualization to meet users’different privacy needs.Scatter plots allow participants to obtain point-level projection in-formation for all data sets.However,aggregation visualization based on density graph supports stricter privacy protection requirements.It hides the point-level relationship of data from other participants and only preserves the density relationship of data.In addition,a set of descriptive views is designed to support visual interactive exploration.(3)This thesis proves the effectiveness of the proposed method through three case studies.The first case is the data purchase scenario of the data market.The data buyer uses this method to analyze the quality of data provider data and the matching degree of data provider data and local data.The second case uses this method to improve the traditional data shar-ing strategy for multiple federated learning users,and better alleviates the problem of non-iid.In the third case,three community hospitals used this method to analyze the data of community health examination,and under-stood the health status of the whole community.After the case was com-pleted,we made a return visit to the participants and recorded and summa-rized the experts’suggestions and opinions.