Unexpected Information Leakage of Differential Privacy Due to the Linear Property of Queries

Differential privacy is a widely accepted concept of preservation, and the Laplace mechanism famous instance differentially private mechanisms used to deal with numerical data. In this paper, we find that differential does not take linear property queries into account, resulting in unexpected information leakage. Specifically, makes it possible divide one query two queries, such as $q(D)=q(D_{1})+q(D_{2})$ if $D=D_{1}\cup D_{2}$ $D_{1}\cap D_{2}=\emptyset $ . If attackers try obtain an answer $q(D)$ , they can only issue but also $q(D_{1})$ calculate $q(D_{2})$ by themselves long know $D_{2}$ Through different divisions query, multiple answers same from mechanisms. However, attackers’ perspective mechanisms’ perspective, total consumed budget are delicately designed. This difference leads leakage because key parameter for controlling amount legally released To demonstrate leakage, present membership inference attack against mechanism. under constraints privacy, propose method obtaining independent identically distributed samples satisfy property. The proposed based on some background knowledge attackers. When sufficient, sufficient number be made unreasonably large. Based obtained samples, hypothesis testing determine whether target record dataset.