Optimal and Differentially Private Data Acquisition: Central and Local Mechanisms
研究了如何设计数据获取机制，在保护用户隐私的同时最小化估计误差，并设计支付规则补偿用户隐私损失，适用于需要从个体收集敏感数据的场景。
The data for many machine learning tasks are owned by individuals who are typically concerned about privacy. Here, the authors study the optimal design of a data acquisition mechanism aimed at learning the mean of a population. This data acquisition scheme includes the design of a payment rule to compensate users for their privacy loss. It also involves selecting an estimator that minimizes estimation error while simultaneously providing privacy guarantees to users in line with their privacy preferences. The authors formulate this problem as a Bayesian mechanism design problem and propose approximately optimal data acquisition mechanisms.