A Probabilistic Hybrid Logic for Sanitized Information Systems

Abstract

As privacy-preserving data publication has received much attention in recent years, a common technique for protecting privacy is to release the data in a sanitized form. To assess the effect of sanitization methods, several data privacy criteria have been proposed. Different privacy criteria can be employed by a data manager to prevent different attacks, since it is unlikely that a single criterion can meet the challenges posed by all possible attacks. Thus, a natural requirement of data management is to have a flexible language for expressing different privacy constraints. Furthermore, the purpose of data analysis is to discover general knowledge from the data. Hence, we also need a formalism to represent the discovered knowledge. The purpose of the paper is to provide such a formal language based on probabilistic hybrid logic, which is a combination of quantitative uncertainty logic and basic hybrid logic with a satisfaction operator. The main contribution of the work is twofold. On one hand, the logic provides a common ground to express and compare existing privacy criteria. On the other hand, the uniform framework can meet the specification needs of combining new criteria as well as existing ones.