Quantile fixed-point estimators for distributional policy evaluation achieve the parametric √n rate, attain the semiparametric efficiency bound for fixed m, remain efficient as m→∞, and admit Berry–Esseen inference.
Distributional Offline Policy Evaluation with Predictive Error Guarantees
1 Pith paper cite this work. Polarity classification is still indexing.
abstract
We study the problem of estimating the distribution of the return of a policy using an offline dataset that is not generated from the policy, i.e., distributional offline policy evaluation (OPE). We propose an algorithm called Fitted Likelihood Estimation (FLE), which conducts a sequence of Maximum Likelihood Estimation (MLE) and has the flexibility of integrating any state-of-the-art probabilistic generative models as long as it can be trained via MLE. FLE can be used for both finite-horizon and infinite-horizon discounted settings where rewards can be multi-dimensional vectors. Our theoretical results show that for both finite-horizon and infinite-horizon discounted settings, FLE can learn distributions that are close to the ground truth under total variation distance and Wasserstein distance, respectively. Our theoretical results hold under the conditions that the offline data covers the test policy's traces and that the supervised learning MLE procedures succeed. Experimentally, we demonstrate the performance of FLE with two generative models, Gaussian mixture models and diffusion models. For the multi-dimensional reward setting, FLE with diffusion models is capable of estimating the complicated distribution of the return of a test policy.
fields
stat.ML 1years
2026 1verdicts
ACCEPT 1representative citing papers
citing papers explorer
-
Statistical Efficiency and Inference of Quantile Distributional Reinforcement Learning
Quantile fixed-point estimators for distributional policy evaluation achieve the parametric √n rate, attain the semiparametric efficiency bound for fixed m, remain efficient as m→∞, and admit Berry–Esseen inference.