Research On Federated Learning Scheme Based On Aggregated Signature

In recent years,driven by the big data and machine learning technology,the data demand of enterprises and researchers has become higher and higher,which is difficult for a single company or organization to meet.However,due to the consideration of all parties for their own data privacy and security,it is difficult to make them share the data with each other.Thus,the problem of data island appears.In response to the problem,federal learning technology emerges as the times require,which completes machine learning training on the basis of protecting the privacy of users’ original data.However,some studies show that the raw data of users is not adequately protected in the current mainstream federated learning scheme.In particular,under the collusion of the aggregation server and some dishonest users,there is a possibility to get the original data by using the exposed gradient data in the federated learning scheme.Besides,the aggregation server also has the possibility to tamper with the aggregation gradient in the scheme process to make users get a wrong model.The series of problems seriously endangers the safety and reliability of federal learning programs.Therefore,the thesis uses literature research method and scientific experiment method to complete our research for these problems.Based on the reading relevant literature and learning the current mainstream academic trend.Then the thesis put forward two federated learning security schemes with high precision and high efficiency.The work of the thesis is mainly for the following two points:(1)For the privacy of federal learning data and effectiveness validation security issues,the thesis based on secret sharing technology and aggregation signature technology puts forward an oriented enterprise federal learning security scheme.Besides,the thesis proves the security of the scheme and gives the performance analysis of the scheme.Finally,the thesis designs a simulation experiment to compare the scheme with other related schemes.The final conclusion shows that the scheme can guarantee the privacy security and the veriability of data with high precision and high efficiency.(2)In view of the disadvantage that the previous scheme without system robustness still has low efficiency,a security scheme for enterprise federal learning with higher efficiency and higher security is proposed.This scheme designs a secure multi-party sharing scheme based on the secret sharing technology,and verifies the data through the aggregated signature technology.Besides this scheme designs a specific protocol to guarantee the robustness of the whole system.Finally,the thesis conducts performance analysis and simulation experiment comparison with related schemes for this scheme.The result of the analysis and experiment proves that this scheme has higher efficiency and safety than other related schemes.