Research On Secure Outsourcing Protocol For AI-assisted Disease Diagnosis And SVM Model Training In Cloud Environmen

In recent years,"cloud computing" has been rapidly developed and widely used due to its advantages of low cost and easy deployment.Especially when computing and storage resource are limited,individuals and organizations can choose cloud server with powerful computing and storage capabilities to complete complex computing tasks.With the vigorous development of artificial intelligence technology,outsourcing data classification and machine learning model training to cloud server has gradually become a common working mode.However,due to the uncontrollability of cloud server,it leads to security problems such as information leakage and falsification of calculation results.Therefore,it is of great research value to design outsourcing algorithms that are safe and efficient without affecting the classification results and training effects.This thesis studies artificial intelligence-assisted disease diagnosis and prediction(AADP)based on hyperplane decision classifier in cloud environment and support vector machine(SVM)model training outsourcing,and propose safe and efficient outsourcing protocol for the above two problems.These two protocols address cloudassisted machine learning applications.The difference is that the AADP protocol focuses on the use of trained machine learning model for disease diagnosis in the cloud server,which belongs to the machine learning classification stage.The SVM training outsourcing protocol uses the cloud server to train the SVM model,which belongs to the training stage of the machine learning model.Now,two protocols are introduced.(1)With the vigorous development of machine learning technology,cloud-assisted AADP protocol has attracted the attention of many scholars as a hot topic.However,the privacy of user data,the intellectual property rights of machine learning model,and the uncontrollability of cloud server bring security challenges to this promising computing model.To solve this problem,this paper innovatively designs a four-party framework composed of users,third-party testing institution,artificial intelligence doctor and cloud server.Under this framework,two efficient and secure AADP protocols are designed based on different security models,and how to deal with the malicious behavior of cloud server is also considered in the protocol.The AADP protocol achieves the goals of security and high efficiency by using secure hash function,Householder transformation,random translation transformation,and random permutation.(2)This thesis discusses the problem of cloud-assisted SVM model training.In order to solve the privacy-preserving problem in the training process,a tripartite framework composed of classification tester,artificial intelligence doctor and cloud server is designed,and a new privacy-preserving SVM training outsourcing protocol is proposed under this framework.Compared with existing protocols,on the basis of protecting the privacy of training data samples and models,the protocol considers the privacy of the correspondence between training data samples and labels for the first time,and hides the access mode of ciphertext data.At the same time,the protocol uses Householder transformation and random permutation for encryption operations,which makes the protocol secure and efficient.