Please note: This PhD seminar will be given online.
Karl
Knopf, PhD
candidate
David
R.
Cheriton
School
of
Computer
Science
Supervisor: Professor Xi He
Differentially private query answering systems have been designed to support interactive exploration of sensitive data. When using differential privacy, the privacy budget is the only parameter that can control the trade-off between the loss of sensitive information and utility of the query answers. Under a limited privacy budget it is possible that not all queries will be able to be answered as accurately as desired by the data analyst.
Prior work has proposed to relax the privacy guarantees by using policies to specify a smaller set of sensitive information. The policy-aware approach allows for a better utility-privacy trade-off than standard differential privacy. A policy-aware differentially private algorithms need to represent the policy as a graph over the full domain of the datasets. This is often inefficient in an interactive setting where the datasets frequently have high dimensions.
In this talk, I will present dynamic Blowfish privacy, which allows for the dynamic generation of a privacy policy based on the exploration query and the predefined privacy-policies defined at attribute level. This offers the same privacy guarantees as the static privacy policy, while being easier to materialize. As part of our research, we have then combined a module implementing dynamic Blowfish privacy with an accuracy-aware query engine. With minimum performance overhead, this system achieves the dual goals of sensitive data exploration by allowing a data curator to specify privacy guarantees while allowing a data analyst to get accuracy-bounded query answers.
To join this PhD seminar on Zoom, please go to https://us02web.zoom.us/j/83326411204?pwd=Z3dNVUxIK01PMXY3MTlXaHNVckJqdz09.