Research On Privacy Computing And Data Consistency In Distributed Scenario

With the development of mobile devices and Internet technology,the data links between different regions and devices are getting closer.In distributed systems,the two most common types are distributed computing systems and distributed information systems.Among them,distributed computing systems focus on using distributed methods to provide computing power for data;distributed information systems focus on storing and processing data in a distributed manner.Distributed computing breaks the physical boundary between data,separates data ownership and processing rights,and causes serious data privacy leaka.ge problems,which makes privacy-preserving computing issues attract more and more researchers’ attention.In the federated learning framework,each client sends updated model parameters to the server after local training,and then the server aggregates the updated model parameters to get the final overall model.Since the client does not need to upload its input to the cloud server,the client can protect the privacy of client input data to a certain extent.However,updating model parameters may compromise the privacy of input data and many malicious input attacks.The first issue we need to solve is maintaining the consistency of data stored by each node in a distributed information system.Blockchain is a typical distributed data-sharing scenario,which uses the consensus mechanism to make all honest nodes save the same blockchain data view.The existing consensus mechanisms(such as Proof-of-work,Proofof-stake,etc.)do not provide a reasonable and fair distribution scheme for the reward distribution after a miner generates a new block.So,the larger node may monopolize most of the mining reward,resulting in the decentralization trend of the whole blockchain.Starting from the privacy computing problem and the data consistency problem during the distributed data process,this paper solved the security aggregation problem of federated learning and improved the consensus mechanism in the blockchain.The concrete research contents are as follow·Research on secure aggregation Protocol in distributed fog computing.This paper combines the efficient additive secret-sharing scheme with federated learning to design a security aggregation protocol based on fog computing scenarios.With the help of fog nodes in the fog computing environment as intermediate processing units,it provides a part of computing services for input data and assists the cloud server in the final security aggregation.This method can reduce the cloud server communication and computing overhead of the secure aggregation protocol.Aiming at the problem of client disconnection during the training process,this paper also designs a lightweight Request-then-Broadcast method to solve this problem.The simulation experiments in large-scale client scenarios showed that the communication time and computing power of this protocol are improved over the previous secure aggregation protocols.·Research on secure aggregation protocols with malicious clients.Aiming at the situation where a large number of clients do aggregation calculations in a federated learning scenario,this paper designs an efficient and secure aggregation protocol in a fog computing scenario that can prevent clients from entering malicious data.This agreement assumes that the fog node in the middle layer has no motive to conspire due to economic benefits or reputation and relevant laws and regulations.Therefore,with the help of these non-collusion fog nodes,after the client sends the secret shares to the fog nodes,the fog nodes can verify the input data using multiplication triples in the secure multi-party computing protocol and drop the malicious input.The server can finally get the correct aggregation result after the verification.Moreover,considering that only a part of the clients is selected to participate in the training process,this paper also gives two different ways of selecting clients according to the special training requirements of the cloud server.·Research on Improvement of PoS Consensus protocol based on Shapley value.In view of the development trend of monopolization of large nodes in the blockchain,this paper improves the reward distribution mode of the PoS mechanism make use of the income distribution principle of Shapley value in the game theory,which makes the reward distribution of nodes participating in the generation of blocks in the PoS mechanism fairer and reasonable,and improves the reward obtained by the new small nodes participating in mining.This method also can slow down the decentralization trend of blockchain and promote the safe and stable operation of the blockchain system.In addition,the same idea also applied to the Ouroboros protocol,and the reward distribution algorithm is improved so that it still satisfies survival and persistence.