Off-Policy Risk Assessment in Markov Decision Processes
This work addresses risk assessment in reinforcement learning for applications like safety alignment, but it is incremental as it extends prior bandit methods to MDPs with improved estimators.
The paper tackles the problem of high variance in off-policy risk assessment for Markov decision processes by developing a doubly robust estimator for the cumulative distribution function of returns, achieving significantly reduced variance and matching minimax lower bounds in experiments.
Addressing such diverse ends as safety alignment with human preferences, and the efficiency of learning, a growing line of reinforcement learning research focuses on risk functionals that depend on the entire distribution of returns. Recent work on \emph{off-policy risk assessment} (OPRA) for contextual bandits introduced consistent estimators for the target policy's CDF of returns along with finite sample guarantees that extend to (and hold simultaneously over) all risk. In this paper, we lift OPRA to Markov decision processes (MDPs), where importance sampling (IS) CDF estimators suffer high variance on longer trajectories due to small effective sample size. To mitigate these problems, we incorporate model-based estimation to develop the first doubly robust (DR) estimator for the CDF of returns in MDPs. This estimator enjoys significantly less variance and, when the model is well specified, achieves the Cramer-Rao variance lower bound. Moreover, for many risk functionals, the downstream estimates enjoy both lower bias and lower variance. Additionally, we derive the first minimax lower bounds for off-policy CDF and risk estimation, which match our error bounds up to a constant factor. Finally, we demonstrate the precision of our DR CDF estimates experimentally on several different environments.