Scholarlink

Privacy preservation is critical in distributed optimization for systems handling sensitive data, creating a fundamental challenge in balancing convergence and privacy guarantees. To address this challenge, a novel Differentially Private Distributed Gradient Algorithm incorporating memory information (DHB-DP) is proposed. DHB-DP effectively maintains fast convergence while ensuring privacy protection by leveraging historical information. Convergence analysis of the proposed algorithm is presented, and strict \(\epsilon \) -differential privacy is mathematically proven. Finally, the effectiveness and attack resistance of the algorithm are verified through numerical simulations and Deep Leakage from Gradients (DLG) attack experiments.

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

Memory-Accelerated Differentially Private Distributed Optimization: Convergence and Privacy Guarantees

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Pan Ouyang,
  • Xin Yi,
  • Jingwen Yi

摘要

Privacy preservation is critical in distributed optimization for systems handling sensitive data, creating a fundamental challenge in balancing convergence and privacy guarantees. To address this challenge, a novel Differentially Private Distributed Gradient Algorithm incorporating memory information (DHB-DP) is proposed. DHB-DP effectively maintains fast convergence while ensuring privacy protection by leveraging historical information. Convergence analysis of the proposed algorithm is presented, and strict \(\epsilon \) -differential privacy is mathematically proven. Finally, the effectiveness and attack resistance of the algorithm are verified through numerical simulations and Deep Leakage from Gradients (DLG) attack experiments.