A federated learning method and apparatus

By introducing a data synchronization and aggregation mechanism between servers in the federated learning system, the problem of low learning efficiency across devices is solved, enabling clients to obtain full data at any time, improving learning efficiency and accuracy, and avoiding waiting and long-tail effects.

CN115249073BActive Publication Date: 2026-03-24HUAWEI TECH CO LTD
View PDF -1 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2021-04-25
Publication Date
2026-03-24

AI Technical Summary

Technical Problem

In cross-device federated learning, learning efficiency is low due to device heterogeneity, especially the differences in storage, computing and communication hardware specifications, which affect performance.

Method used

By introducing multiple servers into the federated learning system, the aggregation information between the servers is synchronized, allowing the client to obtain the full amount of data during each iteration. The client actively requests the server to access and train the model. The servers transmit and aggregate model update parameters to each other. Counters and timers are used to manage the iteration process to limit the number of participating clients and the time window.

Benefits of technology

It improves the overall efficiency of cross-device federated learning, ensuring that clients can obtain full data whenever they access the server, reducing waiting time, avoiding the long tail effect, and improving the accuracy and efficiency of learning.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN115249073B_ABST
    Figure CN115249073B_ABST
Patent Text Reader

Abstract

The application provides a federated learning method and device, which is used for synchronizing the aggregation information between each server in the federated learning process across devices, so that the data of the server in each round of iterative learning process is kept synchronized, each server has more comprehensive data, and the overall learning efficiency is improved. The method comprises the following steps: a first server receives a request message sent by at least one first client; the first server sends a global model and a training configuration parameter to the at least one first client; the first server receives first model update parameters respectively fed back by the at least one first client; the first server aggregates the first model update parameters to obtain first aggregation information in the current iteration; the first server obtains second aggregation information sent by a second server; and the first server updates the global model stored on the first server based on the first aggregation information and the second aggregation information.
Need to check novelty before this filing date? Find Prior Art