Estimating Risk and Uncertainty in Deep Reinforcement Learning
Machine Learning
2020-09-10 v5 Artificial Intelligence
Machine Learning
Abstract
Reinforcement learning agents are faced with two types of uncertainty. Epistemic uncertainty stems from limited data and is useful for exploration, whereas aleatoric uncertainty arises from stochastic environments and must be accounted for in risk-sensitive applications. We highlight the challenges involved in simultaneously estimating both of them, and propose a framework for disentangling and estimating these uncertainties on learned Q-values. We derive unbiased estimators of these uncertainties and introduce an uncertainty-aware DQN algorithm, which we show exhibits safe learning behavior and outperforms other DQN variants on the MinAtar testbed.
Cite
@article{arxiv.1905.09638,
title = {Estimating Risk and Uncertainty in Deep Reinforcement Learning},
author = {William R. Clements and Bastien Van Delft and Benoît-Marie Robaglia and Reda Bahi Slaoui and Sébastien Toth},
journal= {arXiv preprint arXiv:1905.09638},
year = {2020}
}
Comments
Work presented at the ICML 2020 Workshop on Uncertainty and Robustness in Deep Learning