---
title: "On Mitigating the Utility-Loss in Differentially Private Learning: A New Perspective by a Geometrically Inspired Kernel Approach (Abstract Reprint)"
authors:
  - "Mohit Kumar"
  - "Bernhard A. Moser"
  - "Lukas Fischer"
doi: "10.24963/ijcai.2024/952"
url: "https://ijcai.org/proceedings/2024/952"
pdf: "https://ijcai.org/proceedings/2024/0952.pdf"
venue: "Proceedings of the Thirty-Third International Joint Conference on Artificial Intelligence"
year: 2024
pages: "8480-8480"
track: "Journal Track"
keywords:
  - "Journal Track: Journal Track"
---

# On Mitigating the Utility-Loss in Differentially Private Learning: A New Perspective by a Geometrically Inspired Kernel Approach (Abstract Reprint)

Mohit Kumar, Bernhard A. Moser, Lukas Fischer

*Proceedings of the Thirty-Third International Joint Conference on Artificial Intelligence*, pages 8480-8480. DOI: [10.24963/ijcai.2024/952](https://doi.org/10.24963/ijcai.2024/952)

## 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.

## Full text

## On Mitigating the Utility-Loss in Differentially Private Learning: A New Perspective by a Geometrically Inspired Kernel Approach (Abstract Reprint)

Mohit Kumar1 , Bernhard A. Moser2,3 and Lukas Fischer3

1 University of Rostock

2 Johannes Kepler University Linz

3 Software Competence Center Hagenberg GmbH

mohit.kumar@uni-rostock.de, bernhard.moser@scch.at, lukas.fischer@scch.at

Abstract Reprint. This is an abstract reprint of a journal by [Kumar et al., 2024].

## 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 privacysensitive 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.

## References

1. [Kumar et al., 2024] Mohit Kumar, Bernhard Alois Moser, and Lukas Fischer. On mitigating the utility-loss in dif- ferentially private learning: A new perspective by a geo- metrically inspired kernel approach. J. Artif. Intell. Res., 79:515-567, 2024.