Research On Data Privacy Protection And Secure Data Sharing Methods

With the development of Internet of Things(IoT),vehicular networks,5G(5th generation mobile networks)technologies and their related applica-tions,the ever-growing number of smart devices in the network generate a large amount of data by the minute.The analysis of these network data can greatly improve the quality of applications and the services they offer,especially on the user experience.However,as much as the rises of artificial intelligence enables the analysis and mining of the large quantity of data,the use of personal data faces several challenges.First,the users pay more and more attention to data privacy and security now-a-days,due to the increasing number of data leakage incidents related to big data analytics applications.Second,due to the limita-tions of computing and communication resources in the edge network nodes,it is difficult for end devices to effectively process the large amount of data gen-erated.Moreover,methods through collecting the distributed user data from multi-source on heterogeneous devices,then performing the centralized analy-sis consume much resources and incur serious security risks.The secure and reliable distributed cooperation mechanisms for data sharing and utilization are urgently required.Focusing on solving the data leakage and resource limitation in edge net-works issues,this thesis reports on how to protect the data security and privacy and how to share the edge data effectively from the following three aspects:data leakage detection,data privacy protection in edge computing and secure data sharing in edge networks.The main contributions and innovations of this thesis are summarized as follows:(1)Data leakage detection based on context graph modelsTo deal with the problem of data leakage,we develop a data leakage detec-tion scheme based on data content and its correlations.By mapping the original data to the weighted context graph model,the key features of the original data are extracted,and the data forms are simplified.Based on the proposed context graph model,this thesis focuses on improving the efficiency and accuracy of data leakage detection.To improve the detection efficiency,the machine learn-ing algorithms are used to extract the features of graph models and match the graph models,so as to detect the presents of any sensitive data leakage events.To improve the detection accuracy towards transformed data leakage,a score-based walking algorithm for graph matching is used to analyze the inspected data to avoid data leakage.Compared with existing methods,the proposed context based detection method can effectively improve the detection accuracy and efficiency towards transformed data leakage.(2)Federated learning empowered data privacy protectionIn edge networks,the privacy protection of user data faces several chal-lenges.First,because of the pervasiveness of user data and the limitation of computing and communication resources,it is hard to analyze and process the data effectively.Second,due to the multi-source and heterogeneity of dis-tributed user data,it is hard to balance the trade-off between data privacy protec-tion and data utility.Moreover,in the centralized computing modes,the third-party server incurs many security threats and increases the risk of data leakage.In order to deal with these issues and balance data utility and data privacy pro-tection,we propose a data privacy protection scheme based on federated learn-ing.First,for certain computing tasks in vehicular networks,the distributed user data is mapped to data models by extracting data features through feder-ated learning.By using data models for further analysis and processing,the risk of original data leakage is mitigated,and the privacy of user data is enhanced.Second,the data leakage detection models can be constructed through federated learning,which avoids the transmission of original data.The detection model monitors the network data transmitted in edge networks to prevent data leakage.Thus,the privacy protection of user data is enhanced.Third,in order to protect parameter privacy in federated learning process,we propose the asynchronous federated learning scheme.The proposed method integrates local differential privacy with the local training models,and uses distributed update scheme for global model learning,which protects the privacy of trained model parameters.By using federated learning to build data models,the original data computing is transformed into data model processing.The training process is migrated from the centralized server to distributed users.Thus,data privacy in the edge computing process is considerably enhanced.(3)Blockchain and federated learning empowered secure data sharingIn edge networks such as vehicular networks and IoT,the user concerns to data privacy and the lack of mutual trust between users have become ma-jor obstacles to data sharing.Meanwhile,due to the limited resources and data availability,it is hard for a single user to analyze and mine the data ef-fectively.Thus edge data sharing is of great significance.This thesis focuses on how to protect user data privacy and securely share distributed user data.To effectively improve the data utility under limited resources,we propose a distributed collaborative processing method based on federated learning.How-ever,in edge networks,the quality of trained models and the privacy of model parameters can be hardly guaranteed,which hinders users from participating in the federated learning process.Towards the security and privacy issues in data sharing,we integrate blockchain with federated learning to establish a se-cure and reliable data sharing framework.First,in IoT networks,we propose a blockchain-empowered federated learning architecture for secure edge data sharing.By designing a training quality based consensus protocol to integrate blockchain with federated learning,data privacy is enhanced and sharing sys-tem security is guaranteed.In vehicular networks,due to the dynamic network states and strict latency performance requirements,the efficiency of the pro-posed blockchain-empowered federated learning architecture becomes the bot-tleneck for data sharing.Thus we propose an efficient asynchronous federated learning and hybrid blockchain scheme.Furthermore,we formulate the selec-tion of participating nodes as an optimization problem and solve the problem by using Deep Reinforcement Learning(DRL).The utility of the system is maxi-mized and the efficiency of data sharing is improved.