DiApprox: Differential Privacy-based Online Range Queries Approximation for Multidimensional Data

With the increasing use of software services in daily life, the data collected by service providers is massive and sensitive. Although current big data analytics frameworks provide enormous data processing capacity, obtaining appropriate and private responses to large-scale queries quickly and without revealing sensitive information remains a challenging problem. It is clear that Approximate Query Processing (AQP) achieves faster execution with reasonable accuracy loss and Differential Privacy (DP) is popular for enforcing privacy by noising answers to queries. In this paper, we address the problem of combining AQP and DP in multidimensional data based on range queries. We present our private approximation system called DiApprox which takes into account online sampling to accelerate the execution of range queries and minimizes the noise to be injected into the samples and query results in order to preserve the data privacy. Through empirical evaluation, we show that DiApprox is able to approximate aggregation on large datasets over ×21 times faster than exact execution, with high accuracy.