Estimating the success of re-identifications in incomplete datasets using generative models

By Luc Rocher, Julien M. Hendrickx, Y. de Montjoye. Nature Communications, 2019

706 citations recorded by the source corpus.

While rich medical, behavioral, and socio-demographic data are key to modern data-driven research, their collection and use raise legitimate privacy concerns. Anonymizing datasets through de-identification and sampling before sharing them has been the main tool used to address those concerns. We here propose a generative copula-based method that can accurately estimate the likelihood of a specific person to be correctly re-identified, even in a heavily incomplete dataset. On 210 populations, our method obtains AUC scores for predicting individual uniqueness ranging from 0.84 to 0.97, with low false-discovery rate. Using our model, we find that 99.98% of Americans would be correctly re-identified in any dataset using 15 demographic attributes. Our results suggest that even heavily sampled anonymized datasets are unlikely to satisfy the modern standards for anonymization set forth by GDPR and seriously challenge the technical and legal adequacy of the de-identification release-and-forget model. Anonymization has been the main means of addressing privacy concerns in sharing medical and socio-demographic data. Here, the authors estimate the likelihood that a specific person can be re-identified in heavily incomplete datasets, casting doubt on the adequacy of current anonymization practices.

Subfields: Artificial Intelligence, Safety Research, Sociology and Political Science

View Paper: https://www.semanticscholar.org/paper/f34b6423878090c7422c078a3dd6a3db80fb7532

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Safety Research

In Sociology and Political Science

Citations in the corpus

Multimodal biomedical AI (2022, neutral): “Beyond these considerations, a required step for multimodal AI is the appropriate linking of all data types available in the datasets, which represents another challenge owing to the increasing risk of identification of individuals and regulatory constraints 146 .”
Analyzing Leakage of Personally Identifiable Information in Language Models (2023, background): “Prior work studied the leakage of PII in masked language models [36, 61], large language models [26, 54] and Smart Reply classiﬁcation models [32].”
Large image datasets: A pyrrhic win for computer vision? (2020, background): “Institutions, academia, and industry alike, amass millions of images of people without consent and often for unstated purposes under the guise of anonymization, a claim that is both ephemeral [57, 68] and vacuous [30].”
Shifting machine learning for healthcare from development to deployment and from models to data (2022, background): “Even if privacy protocols are followed, patient characteristics can sometimes be inferred from the ML model and its outputs 47,48 .”
Trustworthy Artificial Intelligence: A Review (2022, neutral): “[164] show that they can be insufficient.”
An Agenda for Open Science in Communication (2020, background): “As a result, nonsignificant results often remain unpublished, creating the so-called “file-drawer problem” (Rosenthal, 1979, p. 638).”
Trustworthy AI: From Principles to Practices (2021, background): “In practical scenarios, the empirical evaluation of the risk of leakage of privacy is usually considered [283, 360].”