Mining Distributed Private Databases Using Random Response Protocols

There is a growing demand for sharing data repositories that often contain personal information across multiple autonomous, possibly untrusted, and private databases. This paper discusses constraints imposed by individual privacy as well as institutional data confidentiality on data mining across multiple databases and presents our initial solutions. We develop a suite of decentralized protocols that aim to effectively anonymize the data for each individual database and compute the query results across databases in a probabilistically secure manner. By relaxing the privacy constraints and accuracy requirement, the protocols achieve efficiency and scalability not offered by traditional multiparty secure computation approaches. Our primary viewpoint is that some approximation is tolerable and even desirable for scalable and robust mining across large, multiparty distributed environment.