Subgraph Counting with Local Node Differential Privacy
Researchers give an optimal local node DP algorithm for counting constant-size subgraphs.
The paper gives a noninteractive (ε,0) local node differential privacy algorithm that counts copies of any public constant-size k-node template in an n-node sensitive graph. Its additive error is O(n^{k-1/2}/ε^{k-1}), matching an Ω(n^{k-1/2}) lower bound for templates whose maximum degree exceeds 1, up to poly(1/ε) factors. Prior LNDP network statistics covered only degree-distribution functions. The lower bound is also the first polynomial-in-n separation between central and local node DP that holds for every input graph.
- Noninteractive (ε,0)-LNDP algorithm counts any public k-node template.
- Additive error is O(n^{k-1/2}/ε^{k-1}) for constant k.
- Matching Ω(n^{k-1/2}) lower bound when template maximum degree exceeds 1.
- First polynomial-n gap between central and local node DP on all graphs.
From the original article
1 min read at arxiv.orgWe study the problem of approximately counting subgraphs under local node differential privacy (LNDP), a recently introduced model that enforces strong privacy guarantees in analyses of distributed social networks.
By Sofya Raskhodnikova · arxiv.org
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.