Secure Data Processing Technology Based On Differential Privacy

With the rapid development of information technologies such as mobile Internet and big data,the amount of data generated by social life has increased dramatically.A significant portion of these data are sensitive data related to individual privacy,such as health,location,consumption,and power data.Cloud computing provides a good platform for the collection,storage,and analysis of these large-scale data,enabling its potential application value to be fully explored.However,when the collected data is outsourced to the public cloud,the holders of the data face a severe risk of privacy leakage.Traditional cryptography can theoretically guarantee the security of data transmission and storage;however,it limits the sharing and deep mining of these data to some extent.How to achieve comprehensive privacy protection in the phases of the aggregation,mining,and release of large-scale data in the cloud environment while guaranteeing its high availability,is a common problem that needs to be solved in the application of big data analysis technology.Differential privacy is a newly proposed privacy protection technology based on data perturbation.It has theoretically guaranteed security and is easy to implement.Actually,it has become a standard in the field of privacy protection.Based on the differential privacy technology and cryptographic components such as fully homomorphic encryption and secure multi-party computation,this paper researches the privacy protection in data processing under the cloud computing environment from the aspects of privacy-preserving data mining and publishing.Corresponding privacy protection framework,algorithm and protocol are designed,and the main results are as follows.1.For the privacy protection issues in different phases of data processing in the cloud computing environment,a general framework for Preserving Multiparty Data Privacy(PMDP for short)in the cloud environment is constructed.The framework is based on multi-key fully homomorphic encryption,dynamic secure multi-party computation protocol,and the samplingaggregation differential privacy mechanism.It can provide full life cycle privacy protection for data proxy storage,computation processing,and result publishing in untrusted cloud environments,protecting input privacy,computation privacy and output privacy simultaneously.Security analysis shows that the framework can achieve the desired security goals in the honest model and the noncolluding semi-honest model.In order to overcome the collusion attack in the semi-honest model,a security-enhanced framework sPMDP is further proposed,which can resist collusion attacks under conditions that at least one party is honest.Performance analysis shows that the proposed framework has advantages in security assurance and all-round function,and is more suitable for secure multi-party data aggregation and publishing.2.For the privacy protection in data mining process after data aggregation,this paper takes the commonly used spectral clustering algorithm for research objects,and a constrained spectral clustering algorithm satisfying differential privacy is designed,which is named DP-CSC.The algorithm is based on CCS-L spectral clustering algorithm and Wishart perturbation mechanism.The Wishart perturbation mechanism can be used to add noise to the sample covariance matrix,and we use it to develop a differentially private version of the Laplacian matrix compression of the CCS-L algorithm.Theoretical analysis and simulation experiments show that by properly selecting the parameters' values,the DP-CSC algorithm can achieve a clustering effect similar to the CCS-L algorithm does with acceptable efficiency while protecting the privacy of the clustering results.3.For privacy and security problems in the process of distributed data aggregation and statistical analysis,a statistical analysis protocol under differential privacy for horizontally distributed data named DDP-SA is designed.Aiming at the security problem that the SM-DDP(secure multi-party distributed differential privacy)protocol cannot resist collusion attacks,the DDP-SA protocol is designed based on homomorphic encryption system,zero-knowledge proof technology and distributed differential privacy model,which can be applied to any statistical model function that allows independent calculation of local statistics.The protocol generates noise based on the design concept of the double-blind protocol,and separates the noise accumulation phase from the aggregation of local statistics.At the same time,it uses the homomorphic encryption system combined with random blinding factor injection to protect the aggregation of local statistics.Comparing it with the SM-DDP protocol,the results show that the DDP-SA is superior in security.Applying the DDP-SA protocol to the linear regression problem shows that it has strong practicality under the premise of privacy protection.Finally,we summarize the main work of our research,and forecasted next step of research work.