Differentially Private Online Aggregation
Abstract
Database operations are often performed in batch mode, i.e. the analyst issuing the query must wait till the database has been processed in its entirety before getting feedback. Batch mode is inadequate for large databases since queries can take several hours to process and often an analyst is satisfied with an approximation. Online aggregation greatly improves user experience and saves resources by providing continuous feedback through running confidence intervals. Further, it provides an interface for users to terminate early and allocate resources elsewhere once a sufficient accuracy level has been achieved. Until now, online aggregation has not been studied in a differentially private setting. In this work, we formulate differentially private online aggregation such that it captures the trade-offs between privacy, accuracy, and usability. Further, we develop a family of differentially private mechanisms, which includes our optimal Gap mechanisms, for answering AVG, COUNT, and SUM queries with WHERE conditions. Also, we develop various optimizations to improve the accuracy of the Gap mechanism and empirically confirm that the Gap mechanisms preform the best overall.
Collections
Cite this version of the work
Harry Sivasubramaniam
(2022).
Differentially Private Online Aggregation. UWSpace.
http://hdl.handle.net/10012/17876
Other formats