pre_thresholding.md

Partition Selection with pre-thresholding

Pre-thresholding is a technique in differentially private partition selection that guarantees that some minimum number of privacy units, n, contribute to every partition in an anonymized dataset. This ensures a partition with fewer than n privacy units is never released. For partitions with n or more contributing privacy units, the probability of releasing the partition increases with the number of privacy units.

This feature combines the thresholding capability of k-anonymity together with the guarantees of $$(\varepsilon,\delta)$$-differential privacy. A minimum threshold may, for instance, be suitable as an extra layer of protection in cases where the partitions themselves are inherently sensitive.

This library includes an implementation of this technique in Java, C++, Go, and Privacy on Beam.