Privacy-preserving Classification And Clustering Method Based On Secure Multi-party Computatio

Machine learning has always been a research hotspot in the industry,and has been widely used in transportation,medical diagnosis and other fields.But training machine learning model requires complex professional knowledge and massive data.Therefore,in order to effectively mine the potential value of massive data and solve the problem of data islands,data can be outsourced to cloud servers for processing.However,outsourcing computing has serious risk of data privacy leakage.In view of this,this dissertation proposes corresponding privacy-preserving scheme for data privacy leakage in the training of classification model and clustering model.(1)Aiming at the problem of data privacy leakage in cloud environment and insufficient accuracy in the privacy-preserving neural network based on homomorphic encryption,a privacy-preserving neural network training scheme is proposed for collaborative dual cloud servers,to achieve the goal of data transmission,computing security and model parameter under the collaborative training process of dual cloud servers.Firstly,in order to avoid using polynomial approximation method to realize nonlinear function such as exponent,and improve the calculation accuracy of nonlinear function,secure multiplication protocol,secure comparison protocol and secure exponent protocol are designed based on Paillier homomorphic encryption and additive secret sharing technology.Furthermore,corresponding secure computing protocols of full connection layer,activation layer,softmax layer and back propagation in neural network are constructed to realize privacy-preserving neural network training scheme based on the designed secure computing protocols.Finally,theoretical and security analysis guarantees the correctness and security of framework.The actual performance results show that compared with the existing schemes,the proposed classification scheme has obvious advantages in model accuracy.(2)Aiming at the problem of data privacy leakage in cloud environment and collusion between cloud servers in the process of clustering,an cooperative securely K-prototype clustering scheme against the rational adversary collusion is proposed.Firstly,aiming at the problem that homomorphic encryption does not directly support nonlinear computing,secure comparison protocol,secure equal computing protocol are designed based on homomorphic encryption and additive secret sharing to ensure that the input data and intermediate results are in the form of additive secret share,so as to prevent a single server from having access to the complete data,and to enable accurate computation of nonlinear function.Secondly,secure distance calculation,secure cluster label update,secure cluster center update,clustering label and clustering center reconstruction are realized based on the designed secure computing protocols.Then,according to the game equilibrium theory,incentive-compatible mechanism is designed,and mutual condition contract and report contract are constructed to constrain cloud servers to implement secure computing protocols non-collusively,so as to achieve the consistency between the individual profit maximization and the global profit maximization.Finally,the proposed protocols and smart contracts are analyzed theoretically,and the performance of protocols and the scheme is verified by experiment.The experimental results show that compared with the model accuracy in plaintext environment,the model accuracy loss of the proposed scheme is controlled within 0.22%,which verifies the effectiveness of the scheme.