Research On Secure Machine Learning Based On Homomorphic Encryption

In recent years,with the growth of data resources and computing power,machine learning has developed rapidly and has been applied to various fields,such as image recog-nition,pedestrian detection,aviation supervision and so on.The more data used in the machine learning training phase,the higher the accuracy of the machine learning model in the prediction phase.However,the use of massive data poses challenges to local com-puting resources.Thanks to the application of cloud services,a large number of machine learning training and predictions are outsourced to cloud servers.However,cloud services themselves have frequent data security incidents,which intensifies public concerns about private data leakage.In order to solve the problem of data leakage in machine learning in the cloud service environment,based on vector homomorphic encryption(VHE),we propose different privacy protection machine learning algorithms in specific application scenarios.Based on the existing research,we study the secure support vector machine(SVM)service based on VHE,pedestrian detection under privacy protection,and aircraft location authentication under privacy protection.The main research content and results of this paper include the following aspects:1.Study the secure support vector machine service based on homomorphic encryption.We use an improved vector homomorphic encryption scheme(IVHE)to provide a cloud-based secure SVM service(SSVMS),which can support multiple service subscribers and service providers.SSVMS can resist collusion between multiple entities and can ensure the security of the model and the security of subscriber query data.2.Study privacy preservation pedestrian detection based on VHE.Based on VHE,we proposed a privacy-preserving pedestrian detection(PPPD)scheme.PPPD can effectively extract the features of the histogram of oriented gradients(HOG)on the encrypted image and safely train the pedestrian detection model based on SVM on the kernel matrix.In addition,PPPD can also ensure data privacy and security during the entire pedestrian detection process.3.Study aircraft location authentication under privacy preservation based on VHE.Based on VHE,we proposed an accurate and efficient aircraft location verification(AEALV)scheme.AEALU uses a grid-based k nearest neighbor(kNN)algorithm to predict the aircraft location under privacy preservation.In addition,we proposed a rapid authentication technology for the legitimacy of the aircraft's claimed loca-tion,which can greatly improve the efficiency of aircraft location legality verifica-tion.