Differential Privacy in Machine Learning: A Survey from Symbolic AI to LLMs

TL;DR

This survey comprehensively reviews the evolution and application of differential privacy in machine learning, highlighting foundational concepts, integration methods, and practical evaluation techniques to promote secure AI development.

Contribution

It provides an extensive overview of differential privacy's theoretical foundations and its practical integration into machine learning models, covering recent advances from symbolic AI to large language models.

Findings

Differential privacy effectively limits information leakage in ML models.

Various methods exist for integrating DP into training processes.

Evaluation techniques for DP in ML are well-established.

Abstract

Machine learning models should not reveal particular information that is not otherwise accessible. Differential privacy provides a formal framework to mitigate privacy risks by ensuring that the inclusion or exclusion of any single data point does not significantly alter the output of an algorithm, thus limiting the exposure of private information. This survey reviews the foundational definitions of differential privacy and traces their evolution through key theoretical and applied contributions. It then provides an in-depth examination of how DP has been integrated into machine learning models, analyzing existing proposals and methods to preserve privacy when training ML models. Finally, it describes how DP-based ML techniques can be evaluated in practice. By offering a comprehensive overview of differential privacy in machine learning, this work aims to contribute to the ongoing development of secure and responsible AI systems.