Fine-Grained Manipulation Attacks to Local Differential Privacy Protocols for Range Query

Local Differential Privacy (LDP) enables massive sensitive data collection and analysis without any trusted aggregator, thus having been widely deployed by large corporations. However, recent studies indicate that LDP protocols can be easily disrupted by poisoning or manipulation attacks. In particular, manipulation attackers can leverage injected/corrupted fake users to send falsified data, which induces the aggregator to derive LDP estimates close to some pre-set targets. Nevertheless, such manipulation attacks in existing studies are only focused on basic statistics (e.g., frequency estimation and mean/variance estimation). As an important analytic task, the vulnerability of LDP protocols for range query has not been thoroughly examined, especially against fine-grained manipulations. In this paper, we present a systematic study on the fine-grained manipulation attacks to LDP protocols for range query, considering both categories of the tree-based and grid-based protocols. By exploring the attack surfaces, we propose the general fine-grained manipulation frameworks for both categories with deliberate attacking processes. Based on the frameworks, we also develop specific attack algorithms for the representative LDP protocols for range queries. Finally, we explore a possible defense mechanism for mitigating these attacks. Besides theoretical analysis, we have validated both our attacks and defense through extensive experiments on both synthetic and real-world datasets.