基于知识融合蒸馏的个性化联邦学习方法和存储介质

By using knowledge fusion distillation in federated learning to generate synthetic datasets and combining them with the knowledge distillation mechanism, the problems of model drift and performance degradation in Non-IID data scenarios are solved, thereby improving the personalized performance and convergence speed of the model while protecting data privacy.

CN120874967BActive Publication Date: 2026-07-17HOHAI UNIV
View PDF 2 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
HOHAI UNIV
Filing Date
2025-06-19
Publication Date
2026-07-17

Smart Images

  • Figure CN120874967B_ABST
    Figure CN120874967B_ABST
Patent Text Reader

Abstract

本发明公开一种基于知识融合蒸馏的个性化联邦学习方法和存储介质，属于个性化联邦学习技术领域，方法包括：服务器预训练并分发扩散模型至各客户端生成本地合成数据集，随后初始化全局模型作为客户端的学生模型进行知识融合蒸馏训练，由上一轮优化的教师模型指导学生模型训练；训练完成后客户端上传学生模型参数至服务器进行联邦聚合更新全局模型，同时利用合成数据增强下一轮教师模型的个性化特征提取能力与可靠性，形成全局协作与本地个性化协同优化的闭环学习框架。本发明能够解决数据异构性导致客户端模型漂移、性能衰减和收敛速率减缓问题。
Need to check novelty before this filing date? Find Prior Art