On Mitigating the Utility-Loss in Differentially Private Learning: A New Perspective by a Geometrically Inspired Kernel Approach
Abstract
Privacy-utility tradeoff remains as one of the fundamental issues of differentially private machine learning. This paper introduces a geometrically inspired kernel-based approach to mitigate the accuracy-loss issue in classification. In this approach, a representation of the affine hull of given data points is learned in Reproducing Kernel Hilbert Spaces (RKHS). This leads to a novel distance measure that hides privacy-sensitive information about individual data points and improves the privacy-utility tradeoff via significantly reducing the risk of membership inference attacks. The effectiveness of the approach is demonstrated through experiments on MNIST dataset, Freiburg groceries dataset, and a real biomedical dataset. It is verified that the approach remains computationally practical. The application of the approach to federated learning is considered and it is observed that the accuracy-loss due to data being distributed is either marginal or not significantly high.
Cite
Text
Kumar et al. "On Mitigating the Utility-Loss in Differentially Private Learning: A New Perspective by a Geometrically Inspired Kernel Approach." Journal of Artificial Intelligence Research, 2024. doi:10.1613/JAIR.1.15071Markdown
[Kumar et al. "On Mitigating the Utility-Loss in Differentially Private Learning: A New Perspective by a Geometrically Inspired Kernel Approach." Journal of Artificial Intelligence Research, 2024.](https://mlanthology.org/jair/2024/kumar2024jair-mitigating/) doi:10.1613/JAIR.1.15071BibTeX
@article{kumar2024jair-mitigating,
title = {{On Mitigating the Utility-Loss in Differentially Private Learning: A New Perspective by a Geometrically Inspired Kernel Approach}},
author = {Kumar, Mohit and Moser, Bernhard Alois and Fischer, Lukas},
journal = {Journal of Artificial Intelligence Research},
year = {2024},
pages = {515-567},
doi = {10.1613/JAIR.1.15071},
volume = {79},
url = {https://mlanthology.org/jair/2024/kumar2024jair-mitigating/}
}