一种高效安全的联邦学习隐私保护方案OA

A high-efficiency federated learning data aggregation privacy protection scheme is proposed to address the issue of low communication efficiency and inadequate security performance caused by differences in device geographical location,network status,storage capacity,computational capability,and parameter exchange during the federal learning process.This scheme reduces the direct interaction frequency between participating clients and the server by grouping clients and selecting group leaders,utilizes gradient compression techniques to reduce the number of parameters uploaded by clients,and introduces a trusted third party and efficient key protocols to encrypt the parameters uploaded by participants,ensuring their privacy security.Additionally,each user can independently verify the aggregated results returned by the server.Secu-rity analysis indicates that this scheme satisfies indistinguishability and data privacy requirements,and experimental results demonstrate high model accuracy and significant advantages in communication overhead.