Privacy Preserving Off-Policy Evaluation
This addresses privacy concerns for applications like medical or financial data in reinforcement learning, though it is incremental as it extends privacy to off-policy evaluation.
The paper tackles the problem of sensitive data leakage in reinforcement learning by introducing the first differentially private approach for off-policy evaluation, showing empirically that it outperforms previous on-policy methods.
Many reinforcement learning applications involve the use of data that is sensitive, such as medical records of patients or financial information. However, most current reinforcement learning methods can leak information contained within the (possibly sensitive) data on which they are trained. To address this problem, we present the first differentially private approach for off-policy evaluation. We provide a theoretical analysis of the privacy-preserving properties of our algorithm and analyze its utility (speed of convergence). After describing some results of this theoretical analysis, we show empirically that our method outperforms previous methods (which are restricted to the on-policy setting).