Differentially Private Policy Evaluation

Borja Balle, Maziar Gomrokchi, Doina Precup

We present the first differentially private algorithms for reinforcement learning, which apply to the task of evaluating a fixed policy. We establish two approaches for achieving d…