Scholarlink

<p>Federated learning (FL) enables collaborative model training across distributed clients while preserving data privacy. Recent studies have focused on enhancing FL performance by transferring knowledge, which includes feature representations, fine-grained model parameters, and network architectures. From the aspects of methods, challenges, and prospects, this paper presents a detailed review of knowledge-driven FL, classifying existing methods into four categories: federated learning based on knowledge distillation (FL-KD), federated learning based on knowledge transfer (FL-KT), federated learning based on knowledge sharing (FL-KS), and other innovative approaches. We analyze the techniques used for knowledge management and discuss key challenges in this field. Additionally, we explore strategies in knowledge-driven FL, including FL-KD, FL-KT, FL-KS, and emerging approaches. We also summarize the commonly used datasets, evaluation metrics, and baselines in this area. Finally, we highlight current challenges and future directions, offering insights into potential advancements in knowledge-driven 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

Knowledge-driven federated learning: A systematic literature review on approaches, challenges, and prospects

  • 期刊论文

  • 发表日期：

    2025年6月12日

查看全文
  • Xiaogang Lin,
  • Xiaoli Zhao,
  • Zilong Yin,
  • Hao Pan,
  • Panzhao Jing,
  • Weishan Li,
  • Kangwei Wang,
  • Yincan Shu

摘要

Federated learning (FL) enables collaborative model training across distributed clients while preserving data privacy. Recent studies have focused on enhancing FL performance by transferring knowledge, which includes feature representations, fine-grained model parameters, and network architectures. From the aspects of methods, challenges, and prospects, this paper presents a detailed review of knowledge-driven FL, classifying existing methods into four categories: federated learning based on knowledge distillation (FL-KD), federated learning based on knowledge transfer (FL-KT), federated learning based on knowledge sharing (FL-KS), and other innovative approaches. We analyze the techniques used for knowledge management and discuss key challenges in this field. Additionally, we explore strategies in knowledge-driven FL, including FL-KD, FL-KT, FL-KS, and emerging approaches. We also summarize the commonly used datasets, evaluation metrics, and baselines in this area. Finally, we highlight current challenges and future directions, offering insights into potential advancements in knowledge-driven FL.