Abstract:

Public and private organizations collect information about individuals, which they may need to share with others for benign purposes. Yet such data sharing raises legitimate privacy concerns. Syntactic data anonymization produces an accurate but imprecise data representation that prevents the inference of an individual’s record and sensitive associations. The ensuing tradeoff between information and privacy calls for anonymization techniques that maximize the preserved information while enforcing a privacy guarantee. Such privacy guarantees require careful modeling themselves.

In this talk, I will outline three contributions to this field: (i) a framework for privacy preservation that builds upon an one-dimensional solution to derive efficient heuristics for multidimensional data using space-filling techniques; (ii) a privacy condition that bounds the increase in an adversary's confidence regarding each piece of sensitive information, and efficient algorithms that achieve this condition; and (iii) a methodology that anonymizes data by recasting values in a heterogeneous manner on a bipartite matching blueprint, thereby preserving up to 40% more information, while providing the same privacy guarantee and resisting adversaries who know the employed algorithms.

After the talk there will be a small reception outside the auditorium. You are all welcome to join.