Please note: This PhD seminar will take place in DC 3301 and online.
Karl Knopf, PhD candidate
David R. Cheriton School of Computer Science
Supervisor: Professor Xi He
Differential privacy (DP) has become the standard approach for preserving the privacy of a sensitive data release. There are many DP solutions for answering multiple linear queries at once, but they look to minimize the total error of all queries, and not the error introduced to individual queries. In practice, there are accuracy requirements on the queries to ensure their usefulness. When the DP guarantee is fixed before analysis, prior work may not find a solution that can meet any of these requirements.
In this seminar, I will discuss solutions that aim to solve a practical objective: maximize the number of queries that meet their accuracy requirement given a fixed DP guarantee. To do this, I will describe a framework we have proposed, and an iterative algorithm designed using this framework to optimize the objective. I will then describe an approach to further improve this algorithm with a data-aware approach that partitions the data domain. I will then show experiments that demonstrate that our algorithm provides better utility on our objective than prior work for a variety of datasets and workloads.
To attend this PhD seminar in person, please go to DC 3301. You can also attend virtually on Zoom.