Personalized federated learning method, electronic device, and computer-readable storage medium

CN115169575BActive Publication Date: 2026-06-19SHENZHEN QIANHAI HUANRONG LIANYI INFORMATION TECHNOLOGY SERVICES CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
SHENZHEN QIANHAI HUANRONG LIANYI INFORMATION TECHNOLOGY SERVICES CO LTD
Filing Date
2022-06-23
Publication Date
2026-06-19

AI Technical Summary

Technical Problem

Existing federated learning methods cannot effectively address the individualized needs of each client, resulting in insufficient model generalization ability, low communication efficiency, and high training costs for newly added clients.

Method used

The local data model structure is split into a representation layer and a personalization layer. The representation layer is used for server-side parameter aggregation, and the personalization layer is used for local learning. The gradient descent algorithm is used for optimization and updating, and the FedAvg algorithm is used for parameter aggregation.

Benefits of technology

It improves the model's generalization ability and communication efficiency, reduces training costs, especially the adaptation cost of new clients, and reduces the amount of parameters transmitted in each iteration.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN115169575B_ABST
    Figure CN115169575B_ABST
Patent Text Reader

Abstract

This invention discloses a personalized federated learning method, an electronic device, and a computer-readable storage medium. The invention splits the local data model structure into two parts: the first part is a representation layer, which is sent to the server for parameter aggregation to learn a highly generalizable shared representation layer; the second part is a personalized layer, which primarily learns the local data features of each client. The first part ensures the generalization of the data model; when a new client joins, it can be fine-tuned based on the existing shared representation and combined with local data to achieve good results with low training costs. The second part ensures client personalization, allowing the model to better fit local data features and achieve better local performance. With the combination of the two parts, only the first part participates in federated training, significantly reducing the number of parameters transmitted in each iteration and improving communication efficiency.
Need to check novelty before this filing date? Find Prior Art