Research On Edge-Intelligence-Oriented Privacy-Preserving Schemes

To deal with the problems of transmission delay and bandwidth necklace in centralized cloud computing,the concept of edge computing was proposed.Further,combined edge computing and artificial intelligence,the edge intelligence is created.By intelligentizing the edge computing devices in the cloud-edge collaborated system,the edge intelligence is deployed closer to the user side and provides stable and convenient intelligent service for the users according to their personalized data.However,during the process,huge amount of sensitive information from the users are involved.The information leakage can seriously damage the personal privacy and even cause substantial economic losses.Therefore,the privacy problems in edge intelligence environments receives growing attentions nowadays.Many researchers have conducted studies on the privacy challenge,but the state-of-arts have not handled the following problems:(1)For the privacy preserving schemes,the risk of privacy leakage from the edge computing devices has not been comprehensive considered and precautionary measures against them to disclose the privacy are not sufficient.Besides,seldom the privacy preserving edge intelligence service scheme;(2)The current privacy preserving authentication schemes largely only concern about the identity privacy,but the real time information contained in authentication request is not well protected.Aiming at the problems above,two scheme are proposed,including an edge intelligence oriented privacy preserving machine learning prediction service scheme and a privacy preserving authentication scheme for edge intelligence transportation system,and the contributions of the thesis are as follows:(1)An edge intelligence oriented verifiable privacy preserving machine learning scheme is proposed,which is constructed by the modified Okamoto-Uchiyama homomorphic encryption.The scheme prevents the whole process of machine learning prediction service from privacy leakage and ensures the integrity of the prediction result.The batch verification in the scheme supports two modes: multiple results from one user and multiple results from multiple users,which improves the efficiency of the verification and makes better utilization of the edge computing devices.Based on the theoretical proofs and simulation evaluations,the scheme has advantages in efficiency and security compared with the related works.(2)A privacy preserving authentication scheme for edge intelligent transportation system is proposed,which supports mutual identity authentication between each two entities in vehicle authentication service and protect the privacy of real-time information.Based on the theoretical proof and security analysis,the scheme has the most security attributes compared to the related works.In simulation experiments on NS-3,the end-to-end delay is less than 5milliseconds and the packet delivery ratio is over 99% in most of the cases,which shows its efficiency and practicability.