Adaptive Differential Privacy And Its Applications

Since the untrusted data analyst abuses data under semi-honest model,which brings privacy leakage of non-interactive and interactive computation models of data.To this end,the provable differential privacy can be used to achieve the trade-off between privacy preserving and data utility.However,there is no unified method to indicate the trade-off between privacy preserving and data utility.Furthermore,taking into account the trade-off of differential privacy under rational model,if data curator obtains the best privacy preserving,then data analyst gets the minimum data utility,vice versa.Thus,the trade-off of differential privacy leads to the extreme phenomenon of privacy leakage or utility disaster under rational model.Moreover,since the protection strategies of data curator and the analysis strategies of data analyst are mutually restrictive,data curator and data analyst can not achieve the best privacy preserving and the maximum data utility,respectively.Thus,considering rational model,it is the best result that data curator achieves expected privacy preserving and data analyst obtains expected data utility.In this thesis,the expected privacy preserving is a preference for privacy budget,and the expected data utility is a preference for data utility.In summary,there are the following problems of researches of differential privacy and its applications:(1)There is no unified method indicating the trade-off between privacy preserving and data utility of differential privacy under semi-honest model;(2)Considering rational model,differential privacy can not achieve expected privacy preserving of data curator,while it can not achieve expected data utility of data analyst;(3)In local data collection of multi-user collaboration,differential privacy leads to the over-protection and insufficient protection.Thus,considering rational model,differential privacy can not be used to achieve expected privacy preserving of users,and it can not be used to achieve expected data utility of the service provider;(4)The existing works only achieve the trade-off between privacy preserving and data utility of genome data by using differential privacy.Moreover,considering rational model in genome data sharing,differential privacy can not be used to achieve expected privacy preserving of patient or data center,while it can not be used to achieve expected data utility of medical center.To address the above problems,firstly,this thesis proposes the privacy-utility monotonicity indicating the trade-off of differential privacy under semi-honest model.Secondly,since the noise generated differential privacy mechanisms are random,this thesis proposes adaptive differential privacy achieving expected privacy preserving and approximate expected data utility under rational model,and it is applied to local data collection of multi-user collaboration.Finally,since genome data is different from other category of data,this thesis proposes adaptive differential privacy of character achieving expected privacy preserving and expected data utility under rational model,which is similar to adaptive differential privacy,and it is applied to genome data sharing.This thesis shows that differential privacy has privacy-utility monotonicity and bounded privacy-utility monotonicity under semi-honest model,and the proposed proposals under rational model,adaptive differential privacy and its mechanisms,and adaptive differential privacy of character and its mechanisms,not only have important theoretical value,but also have important practical significance of engineering applications.The specific research contents of this thesis are as follows.(1)This thesis proposes privacy-utility monotonicity based on computational indistinguishability according to privacy metric and utility metric of this thesis to indicate the trade-off of differential privacy under semi-honest model.Furthermore,building on bounded privacy metric and bounded utility metric of this thesis,this thesis proposes bounded privacy-utility monotonicity indicating the bounded trade-off of differential privacy.Next,this thesis theoretically and experimentally shows several differential privacy mechanisms maintaining privacy-utility monotonicity,while satisfying bounded privacy-utility monotonicity.Finally,this thesis analyzes that privacy-utility monotonicity of differential privacy leads to unilateral trade-off under rational model,which leads to serious problems of privacy leakage or utility disaster.(2)On the basis of research content(1),since differential privacy mechanisms generate random noise,this thesis proposes the definition of adaptive differential privacy achieving expected privacy preserving and approximate expected data utility according to privacy metric and utility metric of this thesis under rational model,and this thesis gives the application model of adaptive differential privacy and corresponding adaptive differential privacy mechanisms.Through theoretical analysis,application example,and experimental analysis,this thesis shows that adaptive differential privacy mechanisms can achieve expected privacy preserving and approximate expected data utility.Furthermore,the performance analysis of this thesis shows that adaptive differential privacy mechanisms can achieve the trade-off between expected privacy preserving,approximate expected data utility,and computational efficiency.(3)This thesis constructs the privacy preserving model of local data collection of multi-user collaboration based on adaptive differential privacy mechanisms of research content(2)under rational model,and proposes the method of multi-user negotiating privacy budget based on heuristic obfuscation to achieve expected privacy preserving.This thesis theoretically and experimentally shows that local data collection model of multi-user collaboration based on adaptive differential privacy can achieve expected privacy preserving and approximate expected data utility.Moreover,the performance analysis of this thesis shows that local data collection model of multi-user collaboration based on adaptive differential privacy can achieve the trade-off between expected privacy preserving,approximate expected data utility,and computational efficiency.(4)Since genome data is different from other category of data,the proposed adaptive differential privacy of research content(2)can not directly be used to privacy preserving of genome data sharing.Thus,this thesis proposes the definition of adaptive differential privacy of character and its mechanisms to achieve expected privacy preserving and expected data utility based on privacy metric and utility metric of this thesis under rational model,which is similar to the definition of adaptive differential privacy and its mechanisms.This thesis theoretically shows that adaptive differential privacy mechanisms of character can achieve expected privacy preserving and expected data utility.Next,this thesis constructs privacy preserving model of genome data sharing based on adaptive differential privacy mechanisms of character under rational model,and this thesis theoretically and experimentally shows that the genome data sharing model based on adaptive differential privacy of character can achieve expected privacy preserving and expected data utility.Moreover,the performance analysis of this thesis demonstrates that the genome data sharing model based on adaptive differential privacy of character can obtain the trade-off between expected privacy preserving,expected data utility,and computational efficiency.