72
0

Differentially Private Set Representations

Abstract

We study the problem of differentially private (DP) mechanisms for representing sets of size kk from a large universe. Our first construction creates (ϵ,δ)(\epsilon,\delta)-DP representations with error probability of 1/(eϵ+1)1/(e^\epsilon + 1) using space at most 1.05kϵlog(e)1.05 k \epsilon \cdot \log(e) bits where the time to construct a representation is O(klog(1/δ))O(k \log(1/\delta)) while decoding time is O(log(1/δ))O(\log(1/\delta)). We also present a second algorithm for pure ϵ\epsilon-DP representations with the same error using space at most kϵlog(e)k \epsilon \cdot \log(e) bits, but requiring large decoding times. Our algorithms match our lower bounds on privacy-utility trade-offs (including constants but ignoring δ\delta factors) and we also present a new space lower bound matching our constructions up to small constant factors. To obtain our results, we design a new approach embedding sets into random linear systems deviating from most prior approaches that inject noise into non-private solutions.

View on arXiv
Comments on this paper