Releasing Private Data for Numerical Queries

Prior work on private data release has only studied counting queries or linear queries, where each tuple in the dataset contributes a value in [0, 1] and a query returns the sum of the values. However, many data analytical tasks involve numerical values that are arbitrary real numbers. In this paper, we present a new mechanism to privatize a dataset 𝐷 for a given set 𝑄 of numerical queries, achieving an error of Õ ( √ 𝑛 • Δ 𝑤 (𝐷)) for each query 𝑤 ∈ 𝑄, where Δ 𝑤 (𝐷) is the maximum contribution of any tuple in 𝐷 queried by 𝑤. This instance-and query-specific error bound not only is theoretically appealing, but also leads to excellent practical performance.