DP-Auditorium:

This presentation emphasizes the importance of auditing algorithms that process sensitive data—an increasingly critical area given growing concerns over user privacy. It highlights that simply removing identifying information, such as names and addresses, is not sufficient to ensure privacy. This has been demonstrated by several past cases in which users were successfully reidentified from seemingly anonymized data, such as the well-known Netflix case in 2008. To address these shortcomings in data protection, the talk introduces the concept of differential privacy, a more robust definition compared to traditional anonymization techniques. This definition aims to minimize the likelihood of successful reidentification attempts by adding controlled noise to the outputs of algorithms that process private information. To validate the effectiveness of the privacy guarantees offered by these algorithms, Mónica and her team developed the DPAuditorium library. This tool allows for the auditing of various mechanisms that implement differential privacy, providing a systematic methodology to assess whether they truly meet their data protection promises. Throughout her talk, she also discusses the many challenges involved in auditing privacy-preserving mechanisms—from technical complexity to limited access to high-quality data. The importance of such audits is further underscored in practical applications like computational biology, where genomic data is used, and location-based services such as Google Maps, where users’ sensitive information must remain protected. Through this work, Mónica contributes to building a safer and more transparent framework for managing sensitive data in an increasingly interconnected and privacy-vulnerable world.