The Research On Privacy Protection Service Mechanism Based On Cloud Intelligent Big Data

With the rapid development of machine learning,more and more data storage and calculations are outsourced to the cloud for processing.However,although the cloud server features large storage space,strong computing power and low price,it is still managed by an unreliable third party,which disables users to guarantee the security of private data before and after model training.Therefore,problems like how to that the user data is not leaked in each link of machine learning,at the same time,the efficiency of machine learning can be improved on the basis of high data confidentiality,the accuracy of data calculation by training nodes are of great important and significance of researches for the development of machine learning.The paper addresses the privacy protection problem of cloud-based big data in machine learning,and proposes a privacy proteceion service mechanism for cloud-based intelligent big data based on cloud platform by using a combination of theories and methods such as distributed framework,homomorphic encryption,differential privacy,Hsah function,public key encryption scheme,provable security theory and cryptographic attack types.It also theoretically analysis and experimentally verifies its key technologies.The main research work is as follows:(1)In order to ensure that users can dynamically provide cloud servers with the data required for machine learning training in real time,this paper proposes an efficient privacy-preserving scheme that supports multiple users by using a partially homomorphic encryption.To enable all users to use their own public key to dynamically encrypt and upload private data at any time.The scheme converts the private data set outsourced by users into an encrypted random data set and uploads it to the cloud server by means of the XOR operator.In addition,the scheme allows the users to use the reserved tags to download the outsourced data set from the system at any time and to verify its integrity,preventing data from being leaked by cloud server and protect users' privacy data.(2)In order to improve the security and efficiency of the data distribution from data centers to the machine learning training nodes in the cloud server,this paper proposes a mechanism for the cloud server to add noise perturbation to the random data set over ciphertext area.In the scheme,the cloud server converts the random data set into a vector,and does a perform addition operation with the noise vector generated by the Laplace distribution in differential privacy,and prevents the internal training nodes of the adversary and data analyst from colluding to steal user data and train submodels through the unsolvable properties of the indefinite equations,which strictly protect the security of private data training.(3)In order to reduce the time overhead of training nodes and improve the computational efficiency of machine learning,this paper further proposes an efficient distributed machine learning framework based on the parallel system.The training task is distributed to different training nodes through parallelization gain.In this scheme,the overall time cost is further reduced on the basis of reducing the complexity of the algorithm,and the efficiency of outsourcing machine learning is improved.After the theoretical analysis of the feasibility,security and computational complexity of the scheme,the simulation experiments are conducted to verify the efficiency and accuracy of the data calculation.The simulation experiment showseomrn that the computational efficiency has also been significantly improved while the solution ensures the user's privacy and security,and that the accuracy of data computation at each node is not reduced by the improved security and computational efficiency.