Scholarlink

<p>Continuous advances in Internet of Things (IoT) technologies, along with the potential of future 6G networks, enable smart cities to leverage distributed AI and deliver intelligent, real-time services while preserving user privacy. However, 6G-enabled smart cities face unique privacy challenges including massive-scale heterogeneous IoT deployments with varying security capabilities, ultra-low latency requirements that may pressure privacy-preserving mechanisms, and diverse urban data sources requiring cross-domain protection. Federated Learning (FL) is a compelling paradigm in this setting, which allows training models in a decentralized manner across diverse edge devices without sharing sensitive raw data. However, traditional aggregation strategies such as federated averaging (FedAvg) perform poorly when faced with non-Independent and Identically Distributed (non-IID) and diverse urban scenarios, translating into slower convergence and lower model accuracy. To address these challenges, in this paper, we present a novel FL method, termed GenAggFL, which utilizes genetic algorithms to optimize the aggregation process in FL. FedAvg uses fixed weights to achieve aggregation; these weights may vary depending on the volume of data for each client, while GenAggFL adjusts aggregation weights dynamically with respect to the quality and relevance of model updates located at clients. Such flexibility enables GenAggFL to well capture the variety and improve the global model accuracy while maintaining FL’s core privacy guarantees. Our empirical results in non-IID settings show that GenAggFL outperforms FedAvg and FedProx, achieving much better accuracy. These findings indicate GenAggFL is an efficient, potential choice to enhance FL in the context of 6G and IoT.</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

Genetic algorithm-driven aggregation for federated learning in 6G-enabled smart cities

  • 期刊论文

  • 发表日期：

    2026年5月2日

查看全文
  • Rached Fouda,
  • Salah Zidi,
  • Najoua Bennaji,
  • Issam Zidi

摘要

Continuous advances in Internet of Things (IoT) technologies, along with the potential of future 6G networks, enable smart cities to leverage distributed AI and deliver intelligent, real-time services while preserving user privacy. However, 6G-enabled smart cities face unique privacy challenges including massive-scale heterogeneous IoT deployments with varying security capabilities, ultra-low latency requirements that may pressure privacy-preserving mechanisms, and diverse urban data sources requiring cross-domain protection. Federated Learning (FL) is a compelling paradigm in this setting, which allows training models in a decentralized manner across diverse edge devices without sharing sensitive raw data. However, traditional aggregation strategies such as federated averaging (FedAvg) perform poorly when faced with non-Independent and Identically Distributed (non-IID) and diverse urban scenarios, translating into slower convergence and lower model accuracy. To address these challenges, in this paper, we present a novel FL method, termed GenAggFL, which utilizes genetic algorithms to optimize the aggregation process in FL. FedAvg uses fixed weights to achieve aggregation; these weights may vary depending on the volume of data for each client, while GenAggFL adjusts aggregation weights dynamically with respect to the quality and relevance of model updates located at clients. Such flexibility enables GenAggFL to well capture the variety and improve the global model accuracy while maintaining FL’s core privacy guarantees. Our empirical results in non-IID settings show that GenAggFL outperforms FedAvg and FedProx, achieving much better accuracy. These findings indicate GenAggFL is an efficient, potential choice to enhance FL in the context of 6G and IoT.