A statistical database provides statistical information about a population, while maintaining the privacy of individuals in the database. A popular interpretation of this statement, due to Dalenius, says that "anything learnable about an individual, given access to the database, can be learned without access to the database." In non-technical terms, we will discuss why any such definition is problematic, and suggest an alternate notion of privacy for statistical databases, differential privacy, that arises naturally from an observation about the impossibility argument.
A thriving research effort has produced high-quality differentially private solutions for a wide range of data analysis tasks. We will try to give a feel for the broad spectrum of things that can be done by accessing information through a privacy-preserving programming interface. Finally, we will touch on some privacy problems arising in the context of behavioral targeting that are not addressed by this approach, and pose some questions about mitigation.
Cynthia Dwork, a theoretical computer scientist, has made fundamental contributions to cryptography, distributed computing, and complexity theory. Her current focus is the development of a mathematically rigorous framework and algorithmic techniques for the privacy-preserving analysis of data. A Distinguished Scientist at Microsoft, Dwork is a recipient of the Edsger W. Dijkstra Prize and a member of the US National Academy of Engineering and the American Academy of Arts and Sciences.