Scholarlink

While preserving the privacy of FL, DP inevitably degrades the utility (i.e., accuracy) of FL due to model perturbations caused by DP noise added to model updates [11]. Existing studies have considered exclusively noise with a persistent root-mean-square amplitude and overlooked an opportunity to adjust the amplitudes to alleviate the adverse effects of the noise. This chapter presents a new DP perturbation mechanism with a time-varying noise amplitude to protect the privacy of FL and retain the capability of adjusting the learning performance.

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

Privacy-Utility Trade-Off in Federated Learning

  • 章节

  • 发表日期：

    2026年

查看全文
  • Kai Li,
  • Xin Yuan,
  • Wei Ni

摘要

While preserving the privacy of FL, DP inevitably degrades the utility (i.e., accuracy) of FL due to model perturbations caused by DP noise added to model updates [11]. Existing studies have considered exclusively noise with a persistent root-mean-square amplitude and overlooked an opportunity to adjust the amplitudes to alleviate the adverse effects of the noise. This chapter presents a new DP perturbation mechanism with a time-varying noise amplitude to protect the privacy of FL and retain the capability of adjusting the learning performance.