Towards an Artifact to Assess Differential Privacy in Microdata Streams

As continued data breaches allow state-level threat actors to assemble expansive dossiers on populations to carry out information warfare objectives, protecting personal privacy in published data sets and internal data stores is increasingly essential to civilian and societal safety. At the same time, the explosion of high-resolution, high-accuracy microdata streams, such as timestamped geolocation coordinates collected simultaneously by hardware platforms, operating systems, and a multitude of on-device applications and sites establishes a layered, highly-correlated pattern of life that can uniquely identify individuals and allow for targeted information warfare actions. Differential privacy (DP) is an advanced but highly effective technique in protecting sensitive data streams. This robust approach preserves privacy in published data sets through additive statistical noise sampled from Gaussian or Laplacian probability distributions. Data sets that contain highly correlated event-based data require specialized techniques to preserve mathematical DP guarantees in microdata streams beyond “user-level” applications available in most off-the-shelf approaches. Because practitioners need more tools to assess the robustness of differentially private outputs in microdata streams, application errors may result in future reidentification and privacy loss for data subjects. This research yields an artifact that can reassociate events in microdata streams when insufficient naive approaches are used. It also serves as a tool for implementers to validate their approaches in highly correlated event data.