THE PRELIMINARY ORAL EXAMINATION FOR THE DEGREE OF Ph.D. IN COMPUTER SCIENCE FOR

                                 Piotr Mardziel

Popular services such as Facebook, Twitter, and Flicker require users to relinquish private information in return for some social benefit. The services turn this information into profit via, among other things, targeted advertisements. The relinquished control over one's data requires the users to trust the service.

Recent proposals call for a reversal of this existing order, letting users keep tighter control over their own data. To retain the utility associated with the traditional services, for both the user and the service, the users can then provide some secure interface to their information. This requires some means of discerning safe requests from insecure ones. Moreover, preserving utility for uncertain use cases may require permissive, yet sound protection mechanisms.

In our recent work cite{mardziel11belief} we described an approach, which we term emph{knowledge-based security} to (a) explicitly model a potential adversary's background knowledge, (b) how it changes as a result of learning aspects of a users private data, and (c) limit it's certainty as a means of privacy protection. Additionally the system has good compositional properties, an adversary's new knowledge is then the new background knowledge for future interactions.

We propose application of knowledge-based security to protection of time-varying data (eg. location). This setting offers an additional challenge: not only is the protection of the data itself relevant, but so is the protection of its dynamic time-varying characteristics. Current approaches fail to consistently and compositionally capture adversary knowledge and how it changes.

We generalized our work to the coalition sharing setting cite{mardziel12knowledge}, in which several distrusting agents want to make mutual use of their secret data, without revealing it beyond some certainty threshold. The work demonstrates the theoretical feasibility of knowledge-based security for coalitions but the setting imposes severe computational challenges and requires some assumptions which might not realistically hold.

The computational challenges can be better addressed by an investigation of the enforcement of knowledge-based security policies via emph{secure multi-party computation}, a method of performing a query over secret data without revealing this data. Among the restrictive assumptions in the coalition setting is the notion that a participant reveals nothing by their choice of program to execute over the shared set of secrets. This assumption does not hold if the participants are rational agents intent on learning the other's secret, as opposed to merely interacting in the coalition for some mutual good. Additionally, sound reasoning about this information flow can be useful for the detection of collusion among participants, as indicated by knowledge colluding agents wouldn't normally possess. The final aspect of our proposal is the investigation of these aspects of knowledge-based security.

Examining Committee:

Dr. Michael Hicks                                -          Chair

Dr. Amol Deshpande                          -          Dept’s Representative

Dr. Jeff Foster                                      -          Committee Member

EVERYBODY IS INVITED TO ATTEND THE PRESENTATION