Achieving differential privacy for k-nearest neighbors based outlier detection by data partitioning
Published in arXiv, 2021
This paper develops a differentially private approach to k-nearest-neighbor outlier detection by separating fitting from classification and reducing sensitivity through grid-based partitioning. The method achieves strong privacy guarantees while retaining near-nonprivate performance on real-world datasets.
Recommended citation: Jens Rauch, Iyiola E. Olatunji, and Megha Khosla. (2021). "Achieving differential privacy for k-nearest neighbors based outlier detection by data partitioning." arXiv.
Download Paper