Scholarlink

<p>To address the critical challenge of limited bandwidth, client selection has emerged as a pivotal strategy for optimizing Federated Learning (FL). Nevertheless, this approach faces significant challenges, such as non-Independent and Identically Distributed (non-IID) data and system heterogeneity, including varying computational power and bandwidth speeds, which affect time-to-accuracy performance in model training. We present a Deadline-Aware Adaptive Client Selection (DACS) algorithm for FL designed to enhance the final model accuracy by selecting clients with higher statistical utility. Furthermore, by determining adaptive deadlines for client selection, the DACS algorithm enhances time-to-accuracy performance. Specifically, we first analyze the influential factors that increase client utility. We then propose an effective approach for determining the duration of each training round. Consequently, the problem of client selection in FL transforms into a multi-arm bandit problem, for which we propose optimal solutions to enhance FL efficiency. The empirical results demonstrate that the DACS method improves time-to-accuracy performance and enhances the final model accuracy in FL.</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

DACS: Deadline-Aware Adaptive Client Selection in federated learning

  • 期刊论文

  • 发表日期：

    2026年7月3日

查看全文
  • Aref Arefnia,
  • Abdolah Chalechale

摘要

To address the critical challenge of limited bandwidth, client selection has emerged as a pivotal strategy for optimizing Federated Learning (FL). Nevertheless, this approach faces significant challenges, such as non-Independent and Identically Distributed (non-IID) data and system heterogeneity, including varying computational power and bandwidth speeds, which affect time-to-accuracy performance in model training. We present a Deadline-Aware Adaptive Client Selection (DACS) algorithm for FL designed to enhance the final model accuracy by selecting clients with higher statistical utility. Furthermore, by determining adaptive deadlines for client selection, the DACS algorithm enhances time-to-accuracy performance. Specifically, we first analyze the influential factors that increase client utility. We then propose an effective approach for determining the duration of each training round. Consequently, the problem of client selection in FL transforms into a multi-arm bandit problem, for which we propose optimal solutions to enhance FL efficiency. The empirical results demonstrate that the DACS method improves time-to-accuracy performance and enhances the final model accuracy in FL.