提示
一种基于双重身份确认的高效聚类联邦学习方法研究与实现
首发时间：2024-06-12
摘要：联邦学习可以使客户端能够在不共享其本地数据的情况下，共同学习一个共享的机器学习模型。然而，在联邦学习中，由于其高度分散的系统架构，会难免地产生数据异构性问题，也称数据非独立同分布问题。一个单一的相关全局模型可能不存在，因此为每个客户端定制个性化的模型是必要的。联邦学习模型的个性化包含两种方法，但是它们都可能无法获取对每个客户端的本地分布具有良好泛化能力的模型。因此，研究者提出了聚类联邦学习方法来优化联邦学习。针对上述提到的问题，本文提出了一种基于双重身份确认的聚类联邦学习方法diccfl。diccfl采用在服务端进行聚类划分和在客户端进行聚类身份确认两种机制来确认客户端的聚类身份。diccfl方法能够改进确定客户端聚类身份的性能，从而提高聚类联邦学习模型和联邦学习系统的综合表现。本研究中所进行的实验结果也表明，本研究所提出的算法性能优于其他现有的聚类联邦学习方法。
关键词：
for information in english, please click here
an efficient cluster federated learning method based on dual cluster identity confirmation
abstract：federated learning (fl) enables a set of clients to collaboratively learn a shared prediction model without sharing their local data. however, in federated learning, the issue of data heterogeneity, also known as non-i.i.d. data, naturally arises due to the highly decentralized system architecture. a single relevant global model may not exist, so having personalized models for each client is necessary.fl personalization comprises two approaches, but both of them can fail to derive a model that generalizes well to the local distributions of each client. thus, clustered fl has been proposed to produce significantly better results. in this work, we propose a novel algorithm, dual cluster identity confirmation of clustered federated learning (diccfl), to address the existing challenges of clustered fl. diccfl uses two mechanisms, cluster division at the server and confirmation of cluster information at the client, to determine the cluster identity of the client. we use diccfl to improve the performance of determining the cluster affiliation of a client, thereby ameliorating the comprehensive performance of the clustered fl model and fl systems. the experimental results prove that our proposed algorithm performs better than other existing clustered fl methods.
keywords：
基金：
论文图表：
引用
导出参考文献
no.****
同行评议
勘误表
一种基于双重身份确认的高效聚类联邦学习方法研究与实现
评论
全部评论