Scholarlink

<p>Federated unlearning (FU) is crucial for enforcing “the right to be forgotten” in federated learning (FL) by removing a participant’s data influence from the global model. Despite growing demand, existing FU methods are predominantly server-driven, leading to high computational costs and limited user control over data revocation. We propose FedSSU, a decentralized FU framework that enables efficient client-side unlearning. By leveraging saliency maps and similarity constraints, FedSSU precisely removes selected data subsets while reducing reliance on centralized computation. Additionally, we introduce FedSSU-Class, a lightweight extension for class-level unlearning, minimizing storage and communication overhead. Experiments on benchmark datasets show that FedSSU accelerates unlearning by up to 70–90<InlineEquation ID="IEq1"> <InlineMediaObject> <ImageObject Color="BlackWhite" FileRef="11227_2025_7478_Article_IEq1.gif" Format="GIF" Height="13" Rendition="HTML" Resolution="72" Type="Linedraw" Width="19" /> </InlineMediaObject> <EquationSource Format="TEX">\(\times\)</EquationSource> <EquationSource Format="MATHML"><math> <mo>×</mo> </math></EquationSource> </InlineEquation> compared to full retraining while preserving model utility and achieving comparable effectiveness. These results establish FedSSU as a scalable and practical solution for privacy-preserving unlearning in FL systems.</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

FedSSU: flexible and efficient decentralized unlearning for federated learning

  • 期刊论文

  • 发表日期：

    2025年6月7日

查看全文
  • Yuhe Leng,
  • Lei Xu,
  • Jianghua Liu,
  • Xiaojun Zhang,
  • Lin Mei,
  • Youyang Qu,
  • Chungen Xu

摘要

Federated unlearning (FU) is crucial for enforcing “the right to be forgotten” in federated learning (FL) by removing a participant’s data influence from the global model. Despite growing demand, existing FU methods are predominantly server-driven, leading to high computational costs and limited user control over data revocation. We propose FedSSU, a decentralized FU framework that enables efficient client-side unlearning. By leveraging saliency maps and similarity constraints, FedSSU precisely removes selected data subsets while reducing reliance on centralized computation. Additionally, we introduce FedSSU-Class, a lightweight extension for class-level unlearning, minimizing storage and communication overhead. Experiments on benchmark datasets show that FedSSU accelerates unlearning by up to 70–90 \(\times\) × compared to full retraining while preserving model utility and achieving comparable effectiveness. These results establish FedSSU as a scalable and practical solution for privacy-preserving unlearning in FL systems.