Abstract

We lay theoretical foundations for new database release mechanisms that allow third-parties to con struct consistent estimators of population statistics, while ensuring that the privacy of each indi vidual contributing to the database is protected. The proposed framework rests on two main ideas. First, releasing (an estimate of) the kernel mean embedding of the data generating random variable instead of the database itself still allows t parties to construct consistent estimators of a wi class of population statistics. Second, the algorithm can satisfy the definition of differential p vacy by basing the released kernel mean embedding on entirely synthetic data points, while controlling accuracy through the metric available in Reproducing Kernel Hilbert Space. We describe two instantiations of the proposed framework, suitable under different scenarios, and prove theoretical results guaranteeing differential privacy of the resulting algorithms and the consistency of estimators constructed from their outputs.