Federal learning-based client group division method and device

By using neural networks to extract client data features and training model parameters in federated learning, combined with cross-view map matching and mean clustering methods, the problem of poor accuracy of client group division in the existing technology is solved, and more accurate and robust group division is achieved.

CN119719827AActive Publication Date: 2025-03-28SUN YAT SEN UNIV
4 Cites 1 Cited by

Patent Information

Application Number
CN202411692559.5
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-11-25
Publication Date
2025-03-28
Estimated Expiration
2044-11-25

AI Technical Summary

Technical Problem

The existing client group division method for federated learning leads to poor accuracy of group division results, and it is impossible to effectively capture the diversity of client data and the differences in model parameters.

Method used

By obtaining local data sets of multiple clients, using a preset neural network learning model for feature extraction, and outputting the distribution characteristics of the client data set; then, the local data set is used to model the initial federated learning neural network to determine the training federated learning neural network; based on the data set distribution characteristics and model parameters, an optimization objective function is constructed, and the client group division results are generated through the cross-view diagram matching mechanism and mean clustering method.

Benefits of technology

By integrating the similarity between the two perspectives of client data and model parameters, the deviation and uncertainty brought about by a single perspective are reduced, the accuracy of group division results is significantly improved, and the robustness of the model is enhanced.

✦ Generated by Eureka AI based on patent content.
Patent Text Reader

Abstract

The invention discloses a client group division method and device based on federated learning, which are used for solving the technical problem of poor accuracy of a group division result caused by the existing client group division method aiming at federated learning. The method comprises the following steps: inputting an obtained local data set of each client into a preset neural network learning model for feature extraction, and outputting client data set distribution features corresponding to each client; performing model training on the initial federated learning neural network corresponding to each client by adopting the local data set of each client, and determining a training federated learning neural network corresponding to each client; constructing an optimization objective function according to the distribution characteristics of the data set of each client and the model parameters of each training federated learning neural network; solving the optimization objective function based on a cross-view image matching mechanism, and outputting a multi-view fusion similarity matrix; and based on a mean value clustering method, clustering the multi-view fusion similarity matrix to generate a client group division result.
Need to check novelty before this filing date? Find Prior Art