MixCon: Adjusting the Separability of Data Representations for Harder Data Recovery

TL;DR

This paper proposes a method called MixCon that adjusts data representation separability in neural networks to balance data utility and resistance to model inversion attacks, based on theoretical and empirical insights.

Contribution

MixCon introduces a novel objective function to control data separability, enhancing security against inversion attacks without sacrificing utility.

Findings

Identifies sweet-spots of data separability for optimal trade-off

Demonstrates increased resistance to inversion attacks

Maintains data utility while reducing vulnerability

Abstract

To address the issue that deep neural networks (DNNs) are vulnerable to model inversion attacks, we design an objective function, which adjusts the separability of the hidden data representations, as a way to control the trade-off between data utility and vulnerability to inversion attacks. Our method is motivated by the theoretical insights of data separability in neural networking training and results on the hardness of model inversion. Empirically, by adjusting the separability of data representation, we show that there exist sweet-spots for data separability such that it is difficult to recover data during inference while maintaining data utility.