k-anonymity
In plain wordsA rule for anonymizing tables of personal data: blur details such as age and postcode until every person matches at least k others, so no one can be singled out.
A property of a released dataset requiring every record to be indistinguishable from at least k − 1 others on its quasi-identifiers: attributes such as postcode, birth date and sex that do not identify anyone alone but can be linked with outside data to re-identify people. It is achieved by generalizing values (exact age to an age band) and suppressing outliers. k-anonymity resists simple linkage attacks but not homogeneity attacks, in which all k matching records share the same sensitive value, or attacks using background knowledge; refinements such as l-diversity and t-closeness respond to these, and all degrade badly on high-dimensional data. Because it constrains the data rather than the release process and composes poorly across releases, differential privacydifferential privacy, n.: A mathematical promise that the results of analyzing a dataset — or a model trained on it — would look almost the same whether or not any one person's data was included, so little can be learned about that person. has largely replaced it as the formal standard, though it remains common in practice.
After ages were banded by decade and postcodes truncated, the table satisfied 5-anonymity.
Origin
Introduced by Pierangela Samarati and Latanya Sweeney in a 1998 technical report and formalized in Sweeney's 'k-anonymity: a model for protecting privacy' (2002). It was motivated by Sweeney's finding that postcode, birth date and sex single out most Americans, and by her 1997 re-identification of the Massachusetts governor's hospital records.
Related
- differential privacydifferential privacy, n.: A mathematical promise that the results of analyzing a dataset — or a model trained on it — would look almost the same whether or not any one person's data was included, so little can be learned about that person.
- personally identifiable informationpersonally identifiable information, n.: Information that can identify a specific person, such as a name, address, phone number or ID number — or several facts that together point to one person.
- privacy-preserving machine learningprivacy-preserving machine learning, n.: The family of techniques for training and using AI models while keeping the underlying personal or confidential data hidden — from adding careful noise to computing on encrypted data.
- membership inference attackmembership inference attack, n.: An attack that tries to tell whether a particular person's record, or a particular document, was part of an AI model's training data — which can itself reveal private facts.
- differential privacydifferential privacy, n.: A mathematical promise that the results of analyzing a dataset — or a model trained on it — would look almost the same whether or not any one person's data was included, so little can be learned about that person.
Sources
- Sweeney, L. (2002). k-anonymity: a model for protecting privacy. International Journal of Uncertainty, Fuzziness and Knowledge-Based Systems 10(5).
Mentioned in
differential privacydifferential privacy, n.: A mathematical promise that the results of analyzing a dataset — or a model trained on it — would look almost the same whether or not any one person's data was included, so little can be learned about that person.