Input: A Machine Learning Model and a privacy statement

Output: Is the privacy statement true or not for the given model

List of Papers that most closely match the above problem:

Literature database

General Trend:

• The most common/researched method to test the privacy of a Machine Learning model carries out state of the art membership inference attack on the model and uses its results to benchmark the privacy. Some of them create a new privacy metric based on Membership Inference attacks.
• The other common trend seems to be Auditing Differential Privacy. Either using hypothesis testing to find violations or empirically calculating the lower bound value for DP.