Scholarlink

With the spread of smart devices, companies improve their services by collecting and utilizing users’ behavioral data. However, the collected data from the user’s device can create issues concerning the identification of individuals, and hence privacy protection is required. Local Differential Privacy (LDP) is a technique that perturbs the user’s data before sending to the server so that the server is not able to have access to private data. Unfortunately, LDP is vulnerable to a poisoning attack in which a set of malicious users disrupt the estimated statistics by sending crafted data. In 2024, Li et al. showed that fine-grained manipulation of the estimated means is feasible. In this work, we study a new fine-grained attack to a multidimensional data with LDP known as Locally Differentially Private Correlated Key-Value (PCKV) for key-value data. We evaluate the proposed fine-grained PCKV attack from both theoretical and empirical viewpoints.

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

Fine-Grained Data Poisoning Attack to Local Differential Privacy Protocols for Key-Value Data

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Terumi Yaguchi,
  • Hiroaki Kikuchi

摘要

With the spread of smart devices, companies improve their services by collecting and utilizing users’ behavioral data. However, the collected data from the user’s device can create issues concerning the identification of individuals, and hence privacy protection is required. Local Differential Privacy (LDP) is a technique that perturbs the user’s data before sending to the server so that the server is not able to have access to private data. Unfortunately, LDP is vulnerable to a poisoning attack in which a set of malicious users disrupt the estimated statistics by sending crafted data. In 2024, Li et al. showed that fine-grained manipulation of the estimated means is feasible. In this work, we study a new fine-grained attack to a multidimensional data with LDP known as Locally Differentially Private Correlated Key-Value (PCKV) for key-value data. We evaluate the proposed fine-grained PCKV attack from both theoretical and empirical viewpoints.