An efficient communication method for federated learning based on real-time response time balancing

CN115392481BActive Publication Date: 2026-05-26CHONGQING UNIV OF POSTS & TELECOMM
View PDF 3 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
CHONGQING UNIV OF POSTS & TELECOMM
Filing Date
2022-08-15
Publication Date
2026-05-26

AI Technical Summary

Technical Problem

In federated learning, the Straggle problem caused by the heterogeneity of computing power of terminal devices leads to increased communication latency. Existing methods ignore the model training participation of low-response devices, resulting in unbalanced model training and reduced performance.

Method used

By balancing the groups of devices according to their response time in the pre-defined cluster iterative training and constructing a dynamic hierarchical communication architecture based on "cloud server-head node-terminal device", a weighted collaborative training mechanism within the cluster is designed to increase the model training participation and accuracy of low-response devices.

Benefits of technology

It reduces communication latency, improves the accuracy and efficiency of model training, and reduces communication overhead between the terminal and the cloud server.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN115392481B_ABST
    Figure CN115392481B_ABST
Patent Text Reader

Abstract

This invention relates to an efficient communication method for federated learning based on real-time response time balancing, belonging to the field of federated machine learning. First, in pre-defined cluster iterative training, each responding terminal device is evenly distributed into pre-defined computing clusters according to its local model computation time, constructing a hierarchical communication architecture based on a unified "cloud server-head node-terminal device" structure. This indirectly increases the model training participation of low-response devices from a structural perspective. Then, fast-response devices can assist slow-response devices in training. This invention indirectly improves the model training participation of low-response devices by dynamically grouping heterogeneous computing devices, adaptively constructing a hierarchical logical communication architecture, and designing a weighted collaborative training mechanism within the computing clusters. Essentially, it solves the communication latency problem caused by resource heterogeneity in federated machine learning technology, thereby improving the accuracy of the trained model.
Need to check novelty before this filing date? Find Prior Art