Entropic Risk Constrained Soft-Robust Policy Optimization
Machine Learning
2020-06-23 v1 Optimization and Control
Machine Learning
Abstract
Having a perfect model to compute the optimal policy is often infeasible in reinforcement learning. It is important in high-stakes domains to quantify and manage risk induced by model uncertainties. Entropic risk measure is an exponential utility-based convex risk measure that satisfies many reasonable properties. In this paper, we propose an entropic risk constrained policy gradient and actor-critic algorithms that are risk-averse to the model uncertainty. We demonstrate the usefulness of our algorithms on several problem domains.
Keywords
Cite
@article{arxiv.2006.11679,
title = {Entropic Risk Constrained Soft-Robust Policy Optimization},
author = {Reazul Hasan Russel and Bahram Behzadian and Marek Petrik},
journal= {arXiv preprint arXiv:2006.11679},
year = {2020}
}