A model updating method and apparatus, and a communication device

By receiving model update information from child nodes and combining it with historical global models for optimization updates, the low efficiency of global model updates in federated learning is solved, achieving a more efficient model update process.

CN115427969BActive Publication Date: 2026-06-16GUANGDONG OPPO MOBILE TELECOMMUNICATIONS CORP LTD
1 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
GUANGDONG OPPO MOBILE TELECOMMUNICATIONS CORP LTD
Filing Date
2020-05-15
Publication Date
2026-06-16

AI Technical Summary

⚠Technical Problem

In federated learning, the time synchronization and data asynchrony of different child nodes lead to low efficiency in global model updates, especially when there are many nodes, the update rate of the master node is limited by the slowest model update information transmission.

⚗Method used

The master node receives model update information from the child nodes and updates it in conjunction with the historical global model. It optimizes the global model by using weight factors and update step size, taking into account the historical impact of model updates, and avoiding performance degradation caused by asynchronous updates.

🎯Benefits of technology

This improves the efficiency of global model updates in federated learning, avoids performance degradation caused by limited transmission of model update information in some child nodes, and ensures the timeliness and accuracy of model updates.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure 1
    Figure 1
  • Figure 2
    Figure 2
Patent Text Reader

Abstract

The embodiment of the application provides a model updating method and device and communication equipment, the method comprises the following steps: a master node receives first model updating information sent by a slave node, wherein the first model updating information is model updating information of a global model expected by the slave node relative to a first global model; and the master node updates a second global model according to the first model updating information and the first global model, to obtain a third global model.
Need to check novelty before this filing date? Find Prior Art