Design And Implementation Of Federated Learning Management Platform Of Data And Model

With the gradual throwing up of data silos and data privacy issues,traditional centralized machine learning no longer meets our needs,and federated learning has emerged.Today,many open source federated learning frameworks have emerged on the network,such as FATE developed by We Bank,Paddle FL developed by Baidu,and Fed Learner developed by Byte Dance.But they all have several common problems: only users who hold data can construct federated learning models;When users construct federated learning models,the operation is cumborsome and inconvenient;Users are unable to systematically manage the resources they own.These issues have caused many inconveniences and constraints for users when constructing federated learning models,which in turn leads to low efficiency in developing federated learning models.In order to solve the above problems,this thesis conducts secondary development on the basis of the FATE federated learning framework,and designs and implements a federated learning management platform of data and model with horizontal federated learning as the main research direction.The main work content and innovation points of this thesis include:(1)Propose two roles for the use and management of resources in the federated learning process: data provider and model developer,thereby enabling users who do not own datasets to also engage in federated learning under the FATE framework.(2)Add the function of managing relevant resources in the federated learning process.This thesis adds management functions for user datasets,federated learning training processes,and federated learning models on the basis of the FATE framework.This makes federated learning more convenient for users,greatly improving the efficiency of federated learning.(3)Add dataset access tools.This thesis designs an additional data access tool based on the FATE framework,which compensates for the shortcomings of the original dataset import function of FATE and makes it more in line with the platform’s requirements.At the same time,this also makes it more convenient for users to import training datasets.