Computer Science · Artificial Intelligence
Thesis Proposal: Creation and Analysis of Differentially-Private Synthetic Datasets
See on the mapAbstract
Statistical agencies are faced with two conflicting objectives: protecting the privacy of their respondents, and providing researchers and policy makers with useful data. There exists a large body of literature on Statistical Disclosure Limitation (SDL) techniques, describing and evaluating methods for statistical agencies to share collected information to users while satisfying their confidentiality requirements. One proposal is to create and publish synthetic datasets. While methods to create and analyze such datasets have been published, it is still unclear how much privacy protection they offer. In our work, we look at the criterion of differential privacy, which protects every record in a dataset even against an intruder with complete knowledge of all other entries. In this thesis, we plan to create general procedures to generate and analyze synthetic datasets satisfying differential privacy. We also want to clearly measure the trade-off between the loss of information that comes Statisticians working in data collection are faced with two conflicting objectives. On the one hand, their job is to collect and publish useful datasets for analysts to use to design public