NDSS 2025 – On the Robustness Of LDP Protocols For Numerical Attributes Under Data Poisoning Attacks

A research paper from an NDSS 2025 session examines the robustness of local differential privacy protocols for numerical attributes under data poisoning attacks, finding that Square Wave and server-side CFO-based protocols are more resilient than user-side CFO variants, and that hash domain size significantly affects robustness beyond utility. The authors introduce cross-protocol attack gain metrics and propose a zero-shot detection technique leveraging reconstructed distributions, demonstrating improved detection of adversarial manipulation.