Dynamic Anonymization for Marginal Publication

Marginal publication is one of important techniques to help researchers to improve the understanding about correlation between published attributes. However, without careful treatment, it's of high risk of privacy leakage for marginal publications. Solution like ANGEL has been available to eliminate such risks of privacy leakage. But, unfortunately, query accuracy has been paid as the cost for the privacy-safety of ANGEL. To improve the data utility of marginal publication while ensuring privacy-safety, we propose a new technique called dynamic anonymization. We present the detail of the technique and theoretical properties of the proposed approach. Extensive experiments on real data show that our technique allows highly effective data analysis, while offering strong privacy guarantees.