DIMACS/PORTIA Working Group Meeting on Privacy-Preserving Data Mining

March 17, 2004
DIMACS Center, CoRE Building, Rutgers University, Piscataway, NJ

Organizers:

Cynthia Dwork, Microsoft, dwork at microsoft.com

Benny Pinkas, HP Labs, benny.pinkas at hp.com

Rebecca Wright, Stevens Institute of Technology, rwright at cs.stevens-tech.edu

The working group follows a related workshop on March 15 and 16, 2004.

This workshop and working group will bring together researchers and practitioners in cryptography, data mining, and other areas to discuss privacy-preserving data mining. The workshop sessions on March 15 and 16, 2004 will consist of invited talks and discussion. March 17, 2004 will be a "working group" of invited participants to identify and explore approaches that could serve as the basis for more sophisticated algorithms and implementations than presently exist, and to discuss directions for further research and collaboration.

Both the workshop and working group will investigate the construction and exploitation of "private" databases, e.g.

• Merging information from multiple data sets in a consistent, secure, efficient and privacy-preserving manner;
• Sanitizing databases to permit privacy-preserving public study.

In a wide variety of applications it would be useful to be able to gather information from several different data sets. The owners of these data sets may not be willing, or legally able, to share their complete data with each other. The ability to collaborate without revealing information could be instrumental in fostering inter-agency collaboration.

Particular topics of interest include:

• Secure multi-party computation. This is a very general and well-studied paradigm that unfortunately has not been used in practice so far. We will investigate ways to make it more efficient and encourage its deployment.
• Statistical techniques such as data swapping, post-randomization, and perturbation.
• Articulation of different notions and aspects of privacy.
• Tradeoffs between privacy and accuracy.
• Architectures that facilitate private queries by a (semi-trusted) third party.
• Methods for handling different or incompatible formats, and erroneous data. We will investigate ideas from dimension reduction, clustering and searching strategy.
• Additional issues such as insuring the accuracy and reliability of responses, query authentication, logging, auditing, access control and authorization policies.

Next: Call for Participation

Working Group Index

DIMACS Homepage