Scholarlink

<p>Traditional machine learning (ML) requires the aggregation of training data on a central server, which introduces various constraints. Federated learning (FL) emerges as a promising solution for real-world applications where direct aggregation of data is not feasible. It addresses ML challenges by facilitating collaborative learning of a shared model among different clients without the need to exchange their local data with a central server. This decentralized training approach ensures the privacy of clients’ data. Vertical FL (VFL) has gained significance as a crucial learning framework for training models on vertically distributed data. This study proposes an efficient clustering framework for VFL based on <i>k</i>-means with shared attributes among clients (VFCkM). Unlike previous research in vertical federated clustering, where the assumption is that all clients possess identical samples but varied attributes, our approach introduces a new clustering framework incorporating shared attributes among the clients’ data. VFCkM has been comprehensively evaluated against recent existing studies using eight real-world datasets under varying cluster configurations. VFCkM minimizes the average runtime across all real datasets by 85.8, 48.0, and 57.4% comparing to FSDPC, FFCM, and HFDPC, respectively.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

VFCkM: a federated clustering framework based on k-means algorithm for vertically partitioned data with shared attributes

  • 期刊论文

  • 发表日期：

    2025年5月12日

查看全文
  • Oruba Alfawaz,
  • Ali A. El-Moursy,
  • Mohamed Saad,
  • Ahmed M. Khedr

摘要

Traditional machine learning (ML) requires the aggregation of training data on a central server, which introduces various constraints. Federated learning (FL) emerges as a promising solution for real-world applications where direct aggregation of data is not feasible. It addresses ML challenges by facilitating collaborative learning of a shared model among different clients without the need to exchange their local data with a central server. This decentralized training approach ensures the privacy of clients’ data. Vertical FL (VFL) has gained significance as a crucial learning framework for training models on vertically distributed data. This study proposes an efficient clustering framework for VFL based on k-means with shared attributes among clients (VFCkM). Unlike previous research in vertical federated clustering, where the assumption is that all clients possess identical samples but varied attributes, our approach introduces a new clustering framework incorporating shared attributes among the clients’ data. VFCkM has been comprehensively evaluated against recent existing studies using eight real-world datasets under varying cluster configurations. VFCkM minimizes the average runtime across all real datasets by 85.8, 48.0, and 57.4% comparing to FSDPC, FFCM, and HFDPC, respectively.