Achieving privacy and utility in view-based data integration (or: how to answer queries without revealing secrets)

If data owners make a portion of their data available to the public, this can allow users to answer queries that require spanning datasources. But publishing data leads to a risk of revealing information that one wishes to make private.This is the well-known trade-off between the utility of the data publishing mechanism  -- its ability to answer an information need -- and its privacy.

This talk will be about the privacy/utility trade-off where the publishing mechansim is a set of declarative queries -- a set of "database views". There is prior work in the database community analyzing whether a set of views are useful for answering a set of queries, as well as research analyzing whether a view-based publishing mechanism is private. Here we consider the problem of designing views that are both secure and private. We study this both in the presence of background knowledge -- expressed as logical sentences --and in the absence of such knowledge.

Target audience: This will be a simple,  informal talk focusing on the formalization of the problem and some comparison to other (academic) work in data privacy. Only in the last part will there be some technical results, and these will be without details. It will be aimed at researchers in AI, algorithms, database, knowledge representation. and logic. It will not deal with any practical privacy issues!

This is joint work with Efthymia Tsamoura.