The paper presents an investigation into K-norm mechanisms used in differentially private computations. It discusses how these mechanisms can yield more accurate additive noise by using a statistic-specific norm. The paper also highlights the challenges of sampling from the corresponding norm balls, which are d-dimensional convex polytopes. It further suggests that K-norm mechanisms might be an overlooked practical tool for private additive noise. Two problems, sum and ranked vote, are considered to demonstrate the potential of problem-specific K-norm mechanisms.
Publication date: 28 Sep 2023
Project Page: https://arxiv.org/abs/2309.15790v1
Paper: https://arxiv.org/pdf/2309.15790
