Synthetic Data: Revisiting the Privacy-Utility Trade-off

人工合成数据被认为是在各种应用中更好的隐私保护替代方案，而不是传统的数据去识别化。然而，最近的一篇文章对这一观点提出了质疑，指出人工合成数据并没有提供比传统匿名化技术更好的隐私与效用之间的平衡，并且会导致效用损失和高度不可预测的隐私收益。该文章还声称已经发现了 PATEGAN 和 PrivBayes 提供的差分隐私保证中的漏洞。当一项研究声称驳斥或否定之前的发现时，验证和验证该研究非常重要。在我们的工作中，我们分析了文章中描述的隐私游戏的实现，并发现它在高度专业化和受限制的环境中运作，这限制了其发现的适用性。我们的探索还揭示了该游戏未满足有关数据分布的关键前提条件，这导致了对 PATEGAN 和 PrivBayes 提供的差分隐私保证的违反。我们还在更一般和不受限制的环境中进行了隐私效用平衡分析。我们的实验表明，与提供的 k-匿名化实现相比，人工合成数据实现了更有利的隐私效用平衡，从而再次确认了早期的结论。

Synthetic data has been challenged as a better privacy-preserving alternative to traditional anonymization techniques. This paper aims to verify and validate the study that challenges the notion of synthetic data and reaffirm the earlier conclusions on privacy-utility trade-off.

The paper analyzes the implementation of the privacy game described in the article and finds that its findings are limited to a highly specialized and constrained environment. The paper also conducts a privacy-utility trade-off analysis in a more general and unconstrained environment, demonstrating that synthetic data achieves a more favorable privacy-utility trade-off compared to the provided implementation of k-anonymization.

The paper highlights the limitations of the study that challenges the notion of synthetic data and provides a more general and unconstrained analysis of the privacy-utility trade-off. The experimentation is conducted using open-source datasets and code, and the paper suggests further research on improving the quality of synthetic data.

Recent related studies include 'Privacy-Preserving Synthetic Data Generation: A Survey' and 'Differential Privacy: A Survey of Results'.