Vendredi 26 mai 2023
Differential Privacy: Toward a Better Tuning of the Privacy Budget (ε) Based on Risk
Mahboobeh Dorafshanian
Étudiante au doctorat sous la supervision du professeur Mohamed Mejri
Heure: 13h30
Visioconférence: Zoom
Résumé: Companies have key concerns about privacy issues when dealing with big data. Many studies show that privacy preservation models such as Anonymization, k-Anonymity, l-Diversity, and t-Closeness failed in many cases. Differential Privacy techniques can address these issues by adding a random value (noise) to the query result or databases rather than releasing raw data. Measuring the value of this noise (ε) is a controversial topic that is difficult for managers to understand. To the best of our knowledge, a small number of works calculate the value of ε. To this end, this paper provides an upper bound for the privacy budget ε based on a given risk threshold when the Laplace noise is used. The risk is defined as the probability of leaking private information multiplied by the impact of this disclosure. Estimating the impact is a great challenge as well as measuring the privacy budget. This paper shows how databases like UT CID ITAP could be very useful to estimate these kinds of impacts.
