Scholarlink

<p>Federated Learning (FL) is a paradigm for distributed machine learning that enables collaborative model training across decentralized clients without sharing raw data. Despite its advantages, FL faces persistent challenges, including statistical heterogeneity, communication cost, scalability, privacy, and energy constraints. Clustering has emerged as a principled way to organize clients with related data characteristics, supporting personalization and, in suitable settings, more efficient aggregation and scheduling. This survey systematically reviews clustering techniques applied in FL including classical partitioning and hierarchical methods, graph and density-based methods, label-aware and lightweight client grouping, adaptive and reinforcement learning–based clustering for changing populations, and spatio-temporal variants that exploit spatial proximity or mobility patterns, and analyzes their strengths, limitations, and applicability across the Internet of Things (IoT), edge computing, healthcare, wireless, and vehicular systems. We also discuss the relevance of clustered FL for next-generation smart-city infrastructures by enabling scalable and low-latency learning over massive IoT. Finally, we outline open problems and future directions, including rigorous privacy accounting with clustered participation, energy-aware and communication-efficient designs, standardized evaluation protocols, and robustness under evolving data.</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

A systematic survey on clustering in federated learning

  • 期刊论文

  • 发表日期：

    2026年4月20日

查看全文
  • Zouheir Belfeki,
  • Moez Krichen,
  • Salah Zidi

摘要

Federated Learning (FL) is a paradigm for distributed machine learning that enables collaborative model training across decentralized clients without sharing raw data. Despite its advantages, FL faces persistent challenges, including statistical heterogeneity, communication cost, scalability, privacy, and energy constraints. Clustering has emerged as a principled way to organize clients with related data characteristics, supporting personalization and, in suitable settings, more efficient aggregation and scheduling. This survey systematically reviews clustering techniques applied in FL including classical partitioning and hierarchical methods, graph and density-based methods, label-aware and lightweight client grouping, adaptive and reinforcement learning–based clustering for changing populations, and spatio-temporal variants that exploit spatial proximity or mobility patterns, and analyzes their strengths, limitations, and applicability across the Internet of Things (IoT), edge computing, healthcare, wireless, and vehicular systems. We also discuss the relevance of clustered FL for next-generation smart-city infrastructures by enabling scalable and low-latency learning over massive IoT. Finally, we outline open problems and future directions, including rigorous privacy accounting with clustered participation, energy-aware and communication-efficient designs, standardized evaluation protocols, and robustness under evolving data.