On Sketch Based Anonymization That Satisfies Differential Privacy Model

摘要

We consider the problem of developing a user-centric toolkit for anonymizing medical data that uses ε-differential privacy to measure disclosure risk. Our work will use a randomized algorithm, in particular, the application of sketches to achieve differential privacy. Sketch based randomization is a form of multiplicative perturbation that has been proven to work effectively on sparse, high dimensional data. However, a differential privacy model has yet to be defined in order to work with sketches. The goal is to study whether this approach will yield any improvement over previous results in preserving the privacy of data. How much the anonymized data utility is retained will subsequently be evaluated by the usefulness of the published synthetic data for a number of common statistical learning algorithms.