A New Game-based Framework to Systematize the Body of Knowledge on Privacy Inference Risks in Machine Learning
A New Game-based Framework to Systematize the Body of Knowledge on Privacy Inference Risks in Machine Learning
A New Game-based Framework to Systematize the Body of Knowledge on Privacy Inference Risks in Machine Learning
By Mahmoud Ghorbel
January 30, 2023
Summary
State of the art in understanding and mitigating information leakage about training data in machine learning models involves using privacy games to capture threat models and measure the risks of deploying ML models.
There is a growing interest in this area, with researchers working to establish relationships between privacy risks and develop ways to mitigate these risks.
The article presents the first systematization of knowledge about privacy inference risks in ML. It proposes a unified representation of five fundamental privacy risks as games: membership inference, attribute inference, property inference, differential privacy distinguishability, and data reconstruction.
The article also states that the use of privacy games has become prevalent in the literature on machine learning privacy and has been used to support the empirical evaluation of machine learning systems against various threats and to compare the strength of privacy properties and attacks.
It is mentioned that in the future, privacy games can be used to communicate privacy properties, making the threat model and all assumptions about dataset creation and training explicit, and can facilitate discussing privacy goals and guarantees with stakeholders making guidelines and decisions around ML privacy.
Privacy games have been used to capture threat models and measure the risks of deploying ML models.
In the future, privacy games can be used to communicate privacy properties and facilitate discussing privacy goals and guarantees with stakeholders making guidelines and decisions around ML privacy.
Reference
Ghorbel, M. (2023, January 30). A new game-based framework to systematize the body of knowledge on privacy inference risks in Machine Learning. MarkTechPost. Retrieved February 10, 2023, from https://www.marktechpost.com/2023/01/30/a-new-game-based-framework-to-systematize-the-body-of-knowledge-on-privacy-inference-risks-in-machine-learning