A Poisoning-Resilient LDP Schema Leveraging Oblivious Transfer with the Hadamard Transform

In recent years, Local Differential Privacy (LDP) has been actively used to collect and utilize users’ usage history from smart devices with privacy considerations. However, since LDP allows users to add noise by themselves, Cao et al. pointed out that it is vulnerable to poisoning attacks where malicious users can intentionally manipulate data and send it to servers, thereby tamper with the aggregation results. Therefore, this study examines the application of an Oblivious Transfer (OT) protocol to the LDP protocol CMS to improve robustness against poisoning attacks. To address the challenge that the amount of data transmission and processing costs increase in proportion to the length of CMS’s vector, we introduce the Hadamard Count Mean Sketch (HCMS) utilizing the Hadamard transform. The proposed method is experimentally implemented, and its security and efficiency are evaluated using open data.