A Unified Framework for Privacy Preserving Data Clustering

We study the problem of publishing a data table containing personal information, while ensuring individual privacy and maintaining data integrity to the possible extent. One popular technique in literature is through k-anonymization. A release is considered to preserve k-anonymity if the record corresponding to any person cannot be distinguished from that of at least k - 1 other individuals whose information also appears in the release. In order to achieve k-anonymity, we propose an unsupervised learning framework. We further show an instantiation of the framework, which leads to an exemplar-based clustering algorithm for practical applications, and report promising results.