Charles River Analytics Inc., developer of intelligent systems solutions, partnered with Galois, Inc. to integrate our probabilistic modeling expertise into the Defense Advanced Research Projects Agency’s Brandeis program. Under the Brandeis program—which “seeks to develop the technical means to protect the private and proprietary information of individuals and enterprises”—the Galois-led TAMBA team developed technology to evaluate the effectiveness of privacy-aware systems.
“We applied our rich understanding of probabilistic modeling and inference to build a toolkit for evaluating privacy systems. The integration of prior models was a core part of our work on the Brandeis program,” explained Michael Harradon, Scientist at Charles River Analytics and TAMBA teammate. “Alongside Galois, we developed sophisticated adversary models and probabilistic inference approaches to test and evaluate privacy protection systems.”
Some new privacy systems employ encryption and differential privacy algorithms, with the goal of preventing others from inferring sensitive data. Charles River Analytics developed an approach to test these systems—inference algorithms for probabilistic models are used to determine whether the sensitive information under protection can be inferred. The approach incorporates side-channel information and complex domain prior models to evaluate the strength of the system against even sophisticated adversaries.
This probabilistic approach also provided a platform for improving the utility of these systems by enhancing accuracy without reducing privacy.
This material is based upon work supported by the United States Air Force and DARPA under Contract No FA8750-16-C-0022. Any opinions, findings and conclusions or recommendations expressed in this material are those of the author(s) and do not necessarily reflect the views of the United States Air Force and DARPA. Distribution Statement “A” (Approved for Public Release, Distribution Unlimited).