Research On Privacy-preserving Intelligent Algorithms In Cloud Computing

With the rapid development of cloud computing and big data technology,more and more users choose to upload their data to cloud servers.However,since the cloud servers are untrusted,people are increasingly concerned about data privacy.Homomorphic encryption,which supports calculations on ciphertexts without decryption,can protect private data involved in outsourced storage and computation.It is one of the current research hotspots.Aiming at data privacy issues of intelligent algorithms in cloud computing,this paper,based on homomorphic encryption technology,proposes a secure TrAdaboost solution and a secure NFA-based medical diagnosis and treatment solution.The solution solves the problems such as data privacy,complex ciphertext calculations,and multi-user scenarios involved in the solutions.The main research contents of this paper are as follows:(1)Focusing on data privacy issues appeared in transfer learning under cloud computing environments,this paper designs a secure TrAdaboost system by utilizing homomorphic encryption.In view of the system’s calculation requirements,three secure protocols are designed,which have a lower overhead compared with existing protocols.Aiming at the privacy leakage problem of outsourced transfer learning,this paper designs encrypted TrAdaboost training and prediction algorithms based on a dual-server model.While implementing the TrAdaboost training and prediction calculations,this solution prevents leakage of the system’s sensitive data(including training data,request data,trained models,prediction results,and intermediate calculation results)to the cloud or unauthorized users.Through the security and performance analyses,it can be seen that the proposed solution reduces computational and communication overhead and protects data privacy simultaneously.The accuracy of the system’s result is close to that of the algorithm on the plaintext domain(2)Focusing on data privacy of medical diagnosis and treatment computing in cloud computing,this paper utilizes homomorphic encryption and non-deterministic finite automata(NFA)algorithms to design a secure remote medical diagnosis and treatment system.Aiming at the issue of state matching on the ciphertext domain,this solution implements secure state matching calculations and obtains the encrypted matching result.Aiming at the defect that the existing state transition algorithm of NFA requires the user and the server to interact,this solution does not require user participation,only depending on the servers to perform corresponding calculations.Based on the privacy-protection protocols,the system uses the NFA-based encrypted medical model(provided by the medical service provider)to perform automatic medical diagnosis on the patient’s encrypted disease state.And the system calculates the best k therapies for the request patient while protecting the medical model,patient’s disease condition,diagnosis results,and other sensitive information.In addition,to achieve a secure approximate search algorithm,this paper also proposes a privacy-preserving error-tolerant NFA evaluation method and applies it to the field of gene searching.The security analysis proves that this system is secure under the honest-but-curious attack model.At the same time,a large number of experimental analyses of the system show that the calculation and communication overhead of the scheme is reasonable and suitable for practical scenarios.