Research On Differential Privacy Adjustable Gaussian Mechanism For Sensitive Data Gradual Release

Differential Privacy is a late-model privacy protection technology,which can protect privacy by publishing original data with noise.The privacy level determines the accuracy of the published results under differential privacy.Since the previous differential privacy data publishing cannot meet the users' demands with higher accuracy,we proposed the adjustable Gaussian mechanism for sensitive data gradual release under differential privacy.The research contents and results include:(1)Proposed the adjustable Gaussian mechanism.We gave the definition of the adjustable Gaussian mechanism.In the process of statistical data released,we proved that a response 1y that preserves(1 1?,?)-differential privacy has already been published.Then,the privacy level is relaxed to2?,with(2 1? >?),and we will add noise 2V on the basis of noise 1V to publish a more accurate response2 y,while the joint response(1 2y,y)preserves(2 2?,?)-differential privacy.(2)Data release under the adjustable Gaussian mechanism.This thesis used adjustable Gaussian mechanism to reduce the privacy level from 1? to 2?.And a single relaxation of the privacy level was extended to multiple rounds of relaxation.In general,if the original data added Gaussian noise with Markov property(the next added noise is only related to the current noise,has nothing to do with the noise of the past),the data release can achieve multiple relaxation of privacy level,so as to achieve the gradual release of sensitive data under the differential privacy.(3)Evaluation of the adjustable Gaussian mechanism.Based on the loss of privacy under differential privacy,we calculated the privacy loss of the adjustable Laplace mechanism,the Gaussian composite mechanism and the adjustable Gaussian mechanism.Then it is found that the privacy loss is small under the adjustable Gaussian mechanism.Therefore,it is proved that it is advantageous to relax the privacy level under the adjustable Gaussian mechanism.This thesis analyzed that adjustable Gaussian mechanism to relax the privacy level during the experiment.In the process of protecting privacy using this mechanism,we found that the smaller data mean square errors were related to higher data accuracy.The research on the general(?,?)-differential privacy,can adjust the privacy level according to the assessment of user' privacy.We also keep the balance between privacy and data protection.It is conducive to promote the application of differential privacy and boost the development of big data technology.