In the task of parameter estimation, robustness against corrupted samples and differential privacy are two highly desirable properties.
These are also similar to each other in the sense that both require the estimator to not depend too heavily on any one sample.
In this talk we will see how to design an algorithm which is both robust and differentially private. In particular, we will see a method, known as the exponential mechanism, which can be used to turn certain types of robust algorithms into differentially private ones.
