一种基于稀疏适配器的联邦学习方法、系统和介质

By monitoring the performance of the client and server in real time during federated learning and adaptively optimizing the sparse adapter, the problems of low computational efficiency and high cost in existing technologies are solved, achieving more efficient model training and reducing storage and communication costs.

CN117892839BActive Publication Date: 2026-07-17YIJIAN (SHANGHAI) INFORMATION TECH CO LTD
View PDF 2 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
YIJIAN (SHANGHAI) INFORMATION TECH CO LTD
Filing Date
2024-01-15
Publication Date
2026-07-17

Smart Images

  • Figure CN117892839B_ABST
    Figure CN117892839B_ABST
Patent Text Reader

Abstract

本申请提供了一种基于稀疏适配器的联邦学习方法、系统和介质。该方法包括：将所有符合预设筛选要求的存储特征数据对应的用户端作为目标用户端，根据训练需求数据与用户端的存储特征数据选取目标稀疏适配器，对稀疏处理后数据的离散情况进行分析，根据分析结果对目标稀疏适配器进行优化后处理获得一次优化稀疏数据，并输入待训练模型训练获得局部模型，然后上传至服务器进行聚合处理生成聚合模型，对聚合模型进行性能测试，根据测试结果和用户端监控数据对目标稀疏适配器进行二次优化，并进行下一轮模型训练。本申请通过在每一轮训练过程中对稀疏适配器进行自适应优化，以实现提高计算效率、减少存储成本和通信成本的目的。
Need to check novelty before this filing date? Find Prior Art