Robust Probabilistic Inference

Yishay Mansour, Aviad Rubinstein, Moshe Tennenholtz

Abstract: Robust probabilistic inference is an extension of probabilistic inference, where some of the observations are adversarially corrupted. We model it as a zero-sum game between the adversary, who can select a modification rule, and the predictor, who wants to accurately predict the state of nature. Given a black-box access to a Bayesian inference in the classic (adversary-free) setting, our near optimal policy runs in polynomial time in the number of observations and the number of possible modification rules.