Representation Balancing MDPs for Off-Policy Policy Evaluation
Abstract
We study the problem of off-policy policy evaluation (OPPE) in RL. In contrast to prior work, we consider how to estimate both the individual policy value and average policy value accurately. We draw inspiration from recent work in causal reasoning, and propose a new finite sample generalization error bound for value estimates from MDP models. Using this upper bound as an objective, we develop a learning algorithm of an MDP model with a balanced representation, and show that our approach can yield substantially lower MSE in common synthetic benchmarks and a HIV treatment simulation domain.
Cite
Text
Liu et al. "Representation Balancing MDPs for Off-Policy Policy Evaluation." Neural Information Processing Systems, 2018.Markdown
[Liu et al. "Representation Balancing MDPs for Off-Policy Policy Evaluation." Neural Information Processing Systems, 2018.](https://mlanthology.org/neurips/2018/liu2018neurips-representation/)BibTeX
@inproceedings{liu2018neurips-representation,
title = {{Representation Balancing MDPs for Off-Policy Policy Evaluation}},
author = {Liu, Yao and Gottesman, Omer and Raghu, Aniruddh and Komorowski, Matthieu and Faisal, Aldo A and Doshi-Velez, Finale and Brunskill, Emma},
booktitle = {Neural Information Processing Systems},
year = {2018},
pages = {2644-2653},
url = {https://mlanthology.org/neurips/2018/liu2018neurips-representation/}
}