מושגי ליבה
Clipping mechanism optimizes error bounds in shuffle-DP for sum estimation problems.
תקציר
The article discusses the instance-specific error bounds achievable through the clipping mechanism in the shuffle model of differential privacy. It introduces a protocol that partitions the domain to estimate sums efficiently, achieving optimal error rates without splitting privacy budgets. The protocol ensures (ε, δ)-DP and minimizes communication complexity while maintaining accuracy.
The authors propose a one-round solution for high-dimensional sum estimation, improving upon existing two-round protocols. By leveraging domain partitioning and parallel composition, they achieve optimal error bounds with practical implementation.
Overall, the article presents innovative approaches to enhance privacy-preserving sum estimation under differential privacy constraints.
סטטיסטיקה
O(U/ε)
O(maxi xi·log log U/ε)
O(1/ε)
O(√n/ε)
O(Max(D) · log log U/ε)
˜O(Max(D) · log3.5 U p log(1/δ)/ε)
ציטוטים
"The clipping mechanism can achieve an instance-specific error bound depending on Max(D)."
"Domain partitioning lowers the total message complexity of all BaseSumDP instances."
"Our final protocol achieves optimal error rates without splitting privacy budgets."