Using machine learning techniques for de-anonymization

Today we have unprecedented access to datasets bearing huge potential in regard to both business and research. However, beside their unquestionable utility, privacy breaches pose a significant risk to the release of these datasets (e.g., datasets originating from healthcare are good examples), thus service providers must use anonymization techniques to minimize the risk of unwanted disclosure. In this study, we focus on de-anonymization attacks, algorithms that are designed to "reverse" the anonymization process. In particular, we focus on a novel segment of these attacks that involve machine learning to improve robustness and efficiency. Furthermore, we highlight and discuss the similarity between de-anonymization and authentication: how can these algorithms, which are generally perceived as unethical, be used legitimately for security reasons under special constraints.