区块链赋能的异步联邦学习在车联网中实现安全数据共享（英文版）.pdf

4298 IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 69, NO. 4, APRIL 2020 Blockchain Empowered Asynchronous Federated Learning for Secure Data Sharing in Internet of Vehicles Yunlong Lu , Student Member, IEEE, Xiaohong Huang , Member, IEEE, Ke Zhang , Sabita Maharjan , Senior Member, IEEE, and Yan Zhang , Fellow, IEEE AbstractIn Internet of Vehicles (IoV), data sharing among vehicles for collaborative analysis can improve the driving expe- rience and service quality. However, the bandwidth, security and privacy issues hinder data providers from participating in the data sharing process. In addition, due to the intermittent and unreliable communications in IoV, the reliability and efciency of data sharing need to be further enhanced. In this paper, we propose a new architecture based on federated learning to re- lievetransmissionloadandaddressprivacyconcernsofproviders. To enhance the security and reliability of model parameters, we develop a hybrid blockchain architecture which consists of the permissioned blockchain and the local Directed Acyclic Graph (DAG).Moreover,weproposeanasynchronousfederatedlearning schemebyadoptingDeepReinforcementLearning(DRL)fornode selection to improve the efciency. The reliability of shared data is also guaranteed by integrating learned models into blockchain andexecutingatwo-stageverication.Numericalresultsshowthat the proposed data sharing scheme provides both higher learning accuracyandfasterconvergence. Index TermsData sharing, Blockchain, Asynchronous federated learning, Deep reinforcement learning, Internet of Vehicles. I. INTRODUCTION T HErapiddevelopmentofnewcomputingandcommunica- tiontechnologiesin5Gnetworksandbeyondopensuppos- sibilities for advanced vehicular services and applications such as autonomous driving and content delivery, which can yield improveddrivingexperience.Inthiscontext,InternetofVehicles (IoV), a new paradigm that integrates intelligent computing and ManuscriptreceivedSeptember18,2019;revisedDecember9,2019;accepted January16,2020.DateofpublicationFebruary13,2020;dateofcurrentversion April 16, 2020. This work was supported in part by Joint Funds of National Natural Science Foundation of China and Xinjiang under Project U1603261, in part by the Key R ). KeZhangiswiththeSchoolofInformationandCommunicationEngineering, University of Electronic Science and Technology of China, Chengdu 610051, China (e-mail: ). Sabita Maharjan is with the Simula Metropolitan Center for Digital Engi- neeing, 0167 Oslo, Norway, and also with the University of Oslo, 0316 Oslo, Norway (e-mail: sabitasimula.no). Yan Zhang is with University of Oslo, Norway, and also with Simula Metropolitan Center for Digital Engineering, Norway (e-mail: yanzhangieee). Digital Object Identier 10.1109/TVT.2020.2973651 vehicles networking into vehicular networks 1, emerges as a crucial area. In the IoV, a large amount of diverse types of data is constantly generated by the moving vehicles, which includes additional data such as trajectories, trafc information and multimedia data. How to efciently and effectively utilize the massive amount of available data to improve the driving experienceandtoprovideextensivehigh-qualityservicesinIoV, is a problem of paramount importance. Data sharing can mitigate the problem by analyzing and mining data collaboratively for improving the quality of IoV applications. However, in IoV, data sharing faces two crucial challenges. First, the vehicles need to share data efciently despite unreliable inter-vehicle communications. How to im- prove data sharing efciency and reliability needs further and thorough investigation. Second, data providers are getting in- creasinglyconcernedaboutdatasecurityandprivacyissuesthat candiscouragethemfromprovidingthedataavailablewiththem for analysis. How to share data efciently and securely in IoV, therefore, remains an open research problem. Multi-access Edge Computing (MEC) enables edge resource sharing by performing the computing and content storage 2 at the edge of mobile networks, through Device-to-Device communication (D2D) 3. In 4, the authors exploited the DRL-inspired MEC solution to design an optimal edge con- tent caching scheme taking mobility into account in vehicular networks. In 5, the authors addressed the assignment problem of the radio channels of the nodes to promptly construct the dynamic D2D-enabled wireless network by exploiting partially overlapping channels and game theory. Despite these studies focusing on efciency of MEC, further investigations on how to achieve distributed intelligence in MEC are still needed. In this regard, some recent works have exploited edge intelligence for resource sharing in vehicular networks. For example, in 6, the authors adopted Deep Reinforcement Learning (DRL) for designing a data transmission scheduling scheme to minimize transmissioncostsinvehicularnetworks.However,thesecurity problem of resource sharing in distributed scenarios remains unsolved. Recently, blockchain has emerged as a promising technol- ogy to provide distributed secure solutions 7. With the ad- vanced features such as tamper-proof, anonymity and traceabil- ity, blockchain has attracted signicant attention for enhancing security in areas such as Internet of Things (IoT) 8, vehicular 0018-9545 2020 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See ieee/publications/rights/index.html for more information. Authorized licensed use limited to: UNIVERSITY OF OSLO. Downloaded on September 01,2020 at 23:03:30 UTC from IEEE Xplore. Restrictions apply.LU et al.: BLOCKCHAIN EMPOWERED ASYNCHRONOUS FEDERATED LEARNING FOR SECURE DATA SHARING IN IoV 4299 networks 9 and smart grids 10. A series of works have studied leveraging blockchain for data sharing in vehicular networks.Forinstance,in11,theauthorsexploitedconsortium blockchain, which is maintained by Road Side Units (RSUs) for achieving secure data sharing in vehicular edge networks. In 12, the authors designed a blockchain empowered secure data sharing architecture for distributed multiple parties. Al- though use of blockchain offers the possibility to enhance secu- rity of data sharing, it may also adversely affect the efciency aspects due to the need of additional computing and communi- cation for maintaining the blockchain. To improve the efciency and intelligence of blockchain, some studies have explored integrating blockchain with Arti- cialIntelligence(AI).In13,theauthorsproposedasecureand intelligent architecture by integrating AI and blockchain into wireless networks for secure resource sharing in 5G beyond. The authors in 14 improved the blockchain framework by ofoading the computation-intensive mining tasks to nearby MEC nodes. However, while studying the security and privacy issuesofdataandthenetworkintheseintegratedframeworksisa vital research direction, it has witnessed rather limited work. To this end, mitigating the resource cost for integrating blockchain with AI demands closer and further investigation. Federated learning 15, 16 is a promising approach for pri- vacypreservededgeintelligenceindistributedscenarios.While in conventional machine learning, all training data is collected atacentralizedcurator,federatedlearningaddressestheprivacy concerns to a large extent, and also reduces data transmission cost by distributing the training work to users themselves. The local training is executed by users on their own data, which usuallyadoptsthegradientdescentoptimizationalgorithm17. In a federated learning framework, users keep their data with themselvesbutsendtheparameterstotheserverforaggregation. Thisprovidesaparallelschemeforuserstolearnaglobalmodel collaborativelywithrespecttotheirdataprivacy.Thusfederated learning achieves edge intelligence bylearning fromdistributed data in a privacy preserved manner, and exploits blockchain to provide a guaranteed collaboration scheme among untrusted participants for efcient sharing. However,inanIoV,ahighlydynamicenvironmentduetothe mobility of vehicles and unreliable inter-vehicle communica- tions,giverisetoanumberofnewchallengestobesolved.Three aspectsarecrucialinthiscontext.First,thecomputingefciency of blockchain, needs to be improved. Second, the reliability of shareddataneedstobeguaranteed.Theriskthatprovidersshare unqualied data such as malicious and redundant data, should be mitigated. Third, the delay due to federated learning should be reduced to deal with the heterogeneous communication and computing capabilities of the vehicles. In this paper, we address these issues by integrating blockchain and federated learning into IoV for data sharing. We develop a hybrid blockchain - PermiDAG, and improve the federated learning with our node selection algorithm. The contributions of this paper can be summarized as follows. We propose a new hybrid blockchain - PermiDAG, which consists of a main permissioned blockchain maintained by theRSUsandthelocalDirectedAcyclicGraph(DAG)run by the vehicles for efcient data sharing in IoV. We propose an asynchronous federated learning scheme for learning models from the edge data, and further im- prove the efciency of federated learning by selecting the participating nodes to minimize the total cost. Weenhancethereliabilityoflearnedmodelsbyintegrating the learned parameters into blockchain and verifying the qualities of these parameters through two-stage verica- tion. The rest of this paper is organized as follows. Related work is discussed in Section II. The architecture of our blockchain empowered federated learning framework is presented in SectionIII.Wefurtheranalyzethehybridblockchainframework in Section IV. In Section V, we present our DRL enabled optimalnodeselectionalgorithmfortheblockchainempowered federated learning framework in detail. We illustrate numerical results in Section VI. Section VII concludes the paper. II. RELATED WORK The last few years have witnessed the rapid developments in 5 G technologies, in particular the D2D communication. D2D communication has been widely applied in wireless networks such as cellular networks 18, 19, and ad hoc networks 20. In 21, the authors studied the outage probability of D2D communicationandanalyzedthedownlinkoutageprobabilityin amultichannelenvironment.Moreover,theconceptofmultihop D2D communication network systems was proposed in 22, which are applicable to many different wireless technologies withclariedrequirements.TheseworksonD2D-enabledwire- less networks open up promising possibilities for efcient and reliable resource sharing in IoV, where users can share their resources through Vehicle-to-Vehicle (V2V) communication. In D2D-enabled wireless networks, MEC plays a crucial role inparticularforresourcemanagement, datasharingandcontent caching. In 23, the authors proposed a privacy-preserving data sharing scheme with the assistance of fog node, which leverages MEC for data analysis and encryption. In 24, 25, the authors lowered the delay and raised the scalability of MEC by using the proposed intelligent resource optimization scheme which simultaneously considers communication, computation, andmigrationinmobilenetworks.IntegratingMECinvehicular networks, the authors in 26 studied the resource allocation problembycombining loadbalancewithofoading foramulti- user multi-server vehicular edge computing system. The rise of AIbringsnewpossibilitiesforMECtoachieveedgeintelligence. SomeworkshaveutilizedtheadvancedDRLforresourcealloca- tion.Forinstance,in27,theauthorsadoptedadeepQ-learning approach for optimal data transmission scheduling in cognitive vehicular networks to minimize transmission costs. Although these works provide effective MEC schemes to share resources at edge, the security and privacy problem 28 remains a severe threat to edge users. Blockchainhasbeenwidelyusedtoaddresssecurityissuesin distributed scenarios. In terms of resource trading, blockchain plays the role of payment platform, which guarantees the transaction security in Peer-to-Peer (P2P) trading. The authors in 29 proposed a localized P2P electricity trading model for locally buying and selling electricity among electric vehicles. Authorized licensed use limited to: UNIVERSITY OF OSLO. Downloaded on September 01,2020 at 23:03:30 UTC from IEEE Xplore. Restrictions apply.4300 IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 69, NO. 4, APRIL 2020 In30,theauthorsstudiedtheresourcemanagementandpricing problem between the provider and miners by modeling the interaction between them as a Stackelberg game. Blockchain has also been used for securely providing computing services. A blockchain-based fair service-provisioning scheme was pro- posed to address the security problems in untrusted and dis- tributedIIoTscenarios31.Theconsensusmechanism,tamper- proof records, and smart contract technologies in blockchain enablesecuretradingamongparticipantswithoutcentralauthor- ities.Afewworkshavealsoutilizedblockchainfordatasharing in vehicular networks. A data sharing and storage system based on the consortium blockchain (DSSCB) 32 was proposed to address the identity validity and message reliability issues in a vehicular ad-hoc network. In 11, the authors proposed a reputation-based blockchain scheme to ensure security and traceability of data sharing in vehicular networks. Consensus protocols constitute one of the most important components of a blockchain in terms of both the structure and computation and communication requirements. The Proof-of- Work (PoW) protocol has been widely adopted for blockchain for vehicular networks 11, 14. The PoW based mechanism can prevent attacks such as the Distributed Denial-of-Service (DDoS) and Byzantine attacks. However, the PoW mechanism costs much computing resource thus introducing scalability and efciency issues. To improve the efciency of the consen- sus process, Delegated Proof of Stake (DPoS) 33 has also been explored for application in vehicular networks. DPoS uses real-time voting combined with a reputation system to achieve consensus, which can perform the consensus process in a more efcient and democratic mechanism. Recently, the DAG enabled blockchain has emerged as a new technology to improve the efciency and scalability of traditional blockchain. DAG is a directed graph data structure that has a topological order. DAG blockchain uses cumulative PoW instead of the computation-sensitive traditional PoW protocol for achieving efcientconsensus.TheDAG-basedblockchainhasbeenwidely used in IoT 34 and 5 G beyond networks 35. With blockchain to assure the security among untrusted edge users, integration of blockchain and AI has yielded noticeable performance improvement for resource sharing in wireless net- works. In 13, the authors proposed a secure and intelligent resource sharing architecture for next-generation wireless net- works by integrating AI and blockchain into them. Although the integration enhances the performance of security and ef- ciency for resou