English

A Linear Reduction Method for Local Differential Privacy and Log-lift

Information Theory 2021-01-27 v2 math.IT Applications

Abstract

This paper considers the problem of publishing data XX while protecting correlated sensitive information SS. We propose a linear method to generate the sanitized data YY with the same alphabet Y=X\mathcal{Y} = \mathcal{X} that attains local differential privacy (LDP) and log-lift at the same time. It is revealed that both LDP and log-lift are inversely proportional to the statistical distance between conditional probability PYS(xs)P_{Y|S}(x|s) and marginal probability PY(x)P_{Y}(x): the closer the two probabilities are, the more private YY is. Specifying PYS(xs)P_{Y|S}(x|s) that linearly reduces this distance PYS(xs)PY(x)=(1α)PXS(xs)PX(x),s,x|P_{Y|S}(x|s) - P_Y(x)| = (1-\alpha)|P_{X|S}(x|s) - P_X(x)|,\forall s,x for some α(0,1]\alpha \in (0,1], we study the problem of how to generate YY from the original data SS and XX. The Markov randomization/sanitization scheme PYX(xx)=PYS,X(xs,x)P_{Y|X}(x|x') = P_{Y|S,X}(x|s,x') is obtained by solving linear equations. The optimal non-Markov sanitization, the transition probability PYS,X(xs,x)P_{Y|S,X}(x|s,x') that depends on SS, can be determined by maximizing the data utility subject to linear equality constraints. We compute the solution for two linear utility function: the expected distance and total variance distance. It is shown that the non-Markov randomization significantly improves data utility and the marginal probability PX(x)P_X(x) remains the same after the linear sanitization method: PY(x)=PX(x),xXP_Y(x) = P_X(x), \forall x \in \mathcal{X}.

Keywords

Cite

@article{arxiv.2101.09689,
  title  = {A Linear Reduction Method for Local Differential Privacy and Log-lift},
  author = {Ni Ding and Yucheng Liu and Farhad Farokhi},
  journal= {arXiv preprint arXiv:2101.09689},
  year   = {2021}
}
R2 v1 2026-06-23T22:27:52.489Z