Research On Differential Privacy Algorithm Based On Blockchain Technolog

Data can be used for analysis and mining,but direct publishing of data risks privacy leaks.This paper focuses on the research and analysis of Differential Privacy Algorithms based on an Interactive Framework.When Differential Privacy Protection technology is applied to Interactive Framework,the risk of privacy leakage is brought by the fast consumption of the privacy budget and inconsistency of query results.Privacy issues due to tuple-to-tuple correlation in the data set are not considered.The low-level noise disturbance mechanism adds too much noise,which results in low data availability.Solutions are presented separately.The detailed scheme is as follows:(1)Blockchain technology is applied to the Differential Privacy Interactive Framework to save and track the consumption of privacy budgets.The Continuous Query-Differential Privacy Mechanism(CQDPM)is proposed to reuse the noise from previous queries and reduce the consumption of privacy budgets,but the datasets to be published can only support fixed query types.The blockchain system is used to record the consumption of the privacy budget and improve the security and availability of the network structure.The experimental results show that the CQDPM framework has better publishing data availability and lower time cost than existing algorithms.(2)For the current interactive differential privacy protection in which default tuples are independent of each other,this paper analyzes the privacy leak risk in this mode,formalizes the definition of A-dependent differential privacy,quantifies the sensitivity of the Gauss mechanism under this definition,handles the probability dependence between tuples,and designs the differential privacy protection mechanism under the dependent tuples to provide strict privacy assurance.The feasibility of this mechanism is demonstrated by experiments.(3)For the Gaussian mechanism in CQDPM algorithm,the noise calculation method of this mechanism is optimized.This paper finds that noise variance = Θ(1?√),not =Θ(1?).Therefore,based on the Gauss mechanism of Dwork et al.in 2014,this paper use Mill ratio to approximate tail probability.At this time,noise variance can achieve better results,in high-privacy areas( → 0)is optimal,but in low-privacy areas( → ∞)is superior.The experimental results show that the optimized Gaussian mechanism can make the noiseenhanced data more effective.