Hierarchical federal learning method and system in mobile social network scene

By optimizing user combination and edge server association in mobile social network scenarios, KKT conditions are calculated to optimize communication bandwidth and calculation frequency, the problems of large communication overhead and high energy consumption in traditional federated learning methods are solved, and efficient federated learning and low-energy user-side computing are achieved.

CN119729638AActive Publication Date: 2025-03-28SHANGHAI JIAOTONG UNIV
4 Cites 0 Cited by

Patent Information

Application Number
CN202411914505.9
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-12-24
Publication Date
2025-03-28
Estimated Expiration
2044-12-24

AI Technical Summary

Technical Problem

In the mobile social network scenario, traditional federated learning methods have excessive communication overhead and increased energy consumption due to repeated data training and complex social network graph structures, making it difficult to effectively manage users' private data and computing resources.

Method used

By selecting appropriate user combinations and edge server association schemes in the social network graph structure, KKT conditions are calculated to optimize communication bandwidth allocation and user local computing frequency, and a joint optimization algorithm is implemented to schedule users and allocate resources.

Benefits of technology

It significantly reduces the energy consumption on the user side, improves the efficiency and model accuracy of federated learning, and ensures the balance of federated learning performance and energy consumption in social network scenarios.

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

Abstract

The invention provides a hierarchical federal learning method and system in a mobile social network scene, and the method comprises the steps: S1, selecting a user scheme based on N users of a social network graph; selecting a user combination based on the user scheme; n is a constant; s2, associating the users of the user combination to K edge servers so as to obtain a plurality of association schemes; k is a constant; and S3, based on the user and the multiple association schemes, calculating KKT conditions of the association schemes so as to obtain optimal communication bandwidth allocation and local calculation frequency of the user. According to the joint optimization algorithm provided by the invention, the users are efficiently scheduled, the number of preferred users in each round of scheduling is lower than 50% of the number of all users, the transmitting power and the calculation frequency are reasonably distributed, and compared with a traditional baseline algorithm, the energy consumption of the hierarchical federal learning system in a social network scene is at least reduced by 55%.
Need to check novelty before this filing date? Find Prior Art