Research On Multi Technology Fusion Data Trading Method Based On Federated Learning

D ata is the power source to promote the rapid development of the digital econ-omy and an important factor of production.In recent years,data protection constraints have limited data to different enterprises and organizations,forming a number of ”data islands”,which is difficult to play its important value.The emergence of federated learn-ing technology makes it possible for data to be traded or shared among organizations.It can well meet the privacy and security regulations,and has attracted widespread atten-tion from academia and industry.However,it is difficult to meet the multi-dimensional needs of data transaction only by relying on federated learning.This study focuses on exploring data transaction methods that take into account high reliability and efficiency,and can carry more complex transaction scenarios,so as to achieve the purpose of data enabling specific businesses to improve efficiency.This paper summarizes the existing data transaction models and their shortcomings at the present stage,and expounds the problems of high communication cost,unclear incentive allocation scheme and central-ization in the practice of the data transaction mechanism based on federated learning.Through modeling and analyzing the data transaction process and sorting out the main participants and key processes in the transaction process,a multi-technology fusion data transaction method based on federated learning is proposed.This method can reduce the loss of communication time and enhance the reliability of incentive mechanism,make the transaction process data traceable and tamper-proof,and improve the quality of ser-vice.The main work and innovation of this paper are as follows:1.A computer system of training effect evaluation and contribution of data providers is proposed.Aiming at the problems of unclear benefit distribution and lack of evaluation data sources in the data transaction scenario of Federated Learning,the Shapley Value and Trusted Execution Environment technology are introduced to optimize the evaluation process of participants’ contribution in the transaction process,so that the proposed transaction method can adapt to more model training situations and improve the credibility and fairness of benefit distribution.2.A transaction process data storage mechanism is proposed.Aiming at the problem of centralized storage of benefit distribution data and training data in the process of data transaction,the mechanism combines Blockchain,smart contract,distributed file storage and other technologies to achieve decentralized storage of process data,while ensuring that the benefit distribution data can not be tampered with and the model training data can be traceable,and improving the credibility of data trans-action services and the ability of nodes to pursue wrongdoing.3.A model data synchronization algorithm based on tree topology is proposed.Aim-ing at the problem of time loss caused by model data transmission in the process of data transaction based on Federated Learning,the algorithm realizes the parallel transmission of weight parameter data by cutting the model weight file and con-structing the interactive node tree,which reduces the synchronization time com-plexity from linear level to logarithmic level.The experimental results show that compared with the star topology,the communication time consumption is reduced by 34% at most.4.On the basis of the above research,the development and implementation of the multi-technology fusion data trading system is carried out.The function test and performance test show that the system can effectively realize data trading and shar-ing,and verify the effectiveness of the proposed data trading method.