Mean estimation of multidimensional numerical data with local differential privacy
Yunshen Ma, Huixin Xu, Yue Zhang, Hao Henry Wang · 2022
As the localization of differential privacy (DP), local differential privacy (LDP) can provide a higher level of privacy protection by perturbing data at the user’s side, instead of through a third-party data collector. The majority of existing multidimensional data collection methods typically divided the privacy budget equally into k part and use a one-dimensional perturbation algorithm for a random k attributes. This may lead to a waste of the privacy budgets. In this paper, we propose a privacy budget allocation method with weights, where the privacy budget is allocated according to the attribute weights set by the server and the perturbed attribute values are uploaded to the server. The server can collect data with higher availability.