Research On Statistical Estimation Methods For High-dimensional Data Based On Local Differential Privac

The collection and analysis of data can provide valuable information,but it can also lead to privacy concerns.To address this issue,local differential privacy has become the latest privacy standard and has been introduced on platforms such as Chrome,i OS and mac OS.However,statistical estimation of high-dimensional data under the local differential privacy standard may face the problems of large statistical errors and high computational complexity.Therefore,this paper proposes some statistical estimation methods for high-dimensional data satisfying local differential privacy based on existing studies.The main research contents include the following three aspects:(1)In this paper,a high-dimensional data mean estimation method satisfying local difference privacy is proposed.The method solves the problem of large statistical errors when estimating the mean value of high-dimensional data under the local differential privacy criterion.Through experimental evaluation,the statistical error of the method is lower than that of existing methods under different privacy budgets and different dimensions.(2)In this paper,a frequency distribution estimation method for high-dimensional data satisfying local differential privacy is proposed.The method solves the problem of large statistical error when estimating frequency distributions of high-dimensional data under local differential privacy criteria.Through experimental evaluation,the statistical error of the method is lower than that of existing methods under different privacy budgets and different dimensions.(3)In this paper,a joint distribution estimation method for high-dimensional data satisfying local differential privacy is proposed.The method solves the problem of large computational complexity when estimating the joint distribution of high-dimensional data under local differential privacy criteria.Through experimental evaluation,the computational complexity of the method is lower than that of existing methods for different privacy budgets and different dimensions.