A smart cache for differentially private queries

Applied differential privacy

We are designing a smart cache that uses synthetic data to allow users to issue more differentially-private queries before depleting their privacy budget.

Differential privacy (DP) is a privacy definition that provides robust and formal guarantees. It ensures that the removal or addition of a database record affects the outcome of a query result only by a small amount and prevents an adversary from combining DP data with side information to reveal sensitive information. Unfortunately, with differential privacy, each successive query to a dataset increases the privacy loss. When the privacy loss reaches a certain threshold, the privacy "budget" is depleted and no more queries can be issued without a privacy risk.

In this project, we are designing a smart cache that uses a synthetic dataset to allow users to issue more queries to a data analytics system before depleting the budget. Our system answers new queries using synthetic data and real data. We release the answer computed from the fake data and do not consume our budget if the result from the synthetic dataset successfully approximates the real answer. On the other hand, we fall back to the real dataset and consume our privacy budget when the synthetic result is inaccurate. A miss triggers an optimization procedure where we update weights in the synthetic dataset to improve the answers from queries that cause a miss. The smart cache is developed in the context of the LEAP project, where the goal is to use machine learning and privacy advances to improve healthcare.

Additional Information

People: Matheus Stolet

arrow_back Back

Systopia lab is supported by a number of government and industrial sources, including Cisco Systems, the Communications Security Establishment Canada, Intel Research, the National Sciences and Engineering Research Council of Canada (NSERC), Network Appliance, Office of the Privacy Commissioner of Canada, and the National Science Foundation (NSF).