Secure distributed computation of anonymized views of shared databases

We consider the problem of computing efficient anonymizations of partitioned databases. Given a database that is partitioned between several sites, either horizontally or vertically, we devise secure distributed algorithms that allow the different sites to obtain a k-anonymized and £-diverse view of the union of their databases, without disclosing sensitive information. Our algorithms are based on the sequential algorithm [Goldberger and Tassa 2010] that offers anonymizations with utility that is significantly better than other anonymization algorithms, and in particular those that were implemented so far in the distributed setting. Our algorithms can apply to different generalization techniques and utility measures and to any number of sites. While previous distributed algorithms depend on costly cryptographic primitives, the cryptographic assumptions of our solution are surprisingly minimal.