Regret Bounds for Reinforcement Learning with Policy Advice
Abstract
In some reinforcement learning problems an agent may be provided with a set of input policies, perhaps learned from prior experience or provided by advisors. We present a reinforcement learning with policy advice (RLPA) algorithm which leverages this input set and learns to use the best policy in the set for the reinforcement learning task at hand. We prove that RLPA has a sub-linear regret of $\widetilde O(\sqrt{T})$ relative to the best input policy, and that both this regret and its computational complexity are independent of the size of the state and action space. Our empirical simulations support our theoretical analysis. This suggests RLPA may offer significant advantages in large domains where some prior good policies are provided.
Cite
Text
Azar et al. "Regret Bounds for Reinforcement Learning with Policy Advice." European Conference on Machine Learning and Principles and Practice of Knowledge Discovery in Databases, 2013. doi:10.1007/978-3-642-40988-2_7Markdown
[Azar et al. "Regret Bounds for Reinforcement Learning with Policy Advice." European Conference on Machine Learning and Principles and Practice of Knowledge Discovery in Databases, 2013.](https://mlanthology.org/ecmlpkdd/2013/azar2013ecmlpkdd-regret/) doi:10.1007/978-3-642-40988-2_7BibTeX
@inproceedings{azar2013ecmlpkdd-regret,
title = {{Regret Bounds for Reinforcement Learning with Policy Advice}},
author = {Azar, Mohammad Gheshlaghi and Lazaric, Alessandro and Brunskill, Emma},
booktitle = {European Conference on Machine Learning and Principles and Practice of Knowledge Discovery in Databases},
year = {2013},
pages = {97-112},
doi = {10.1007/978-3-642-40988-2_7},
url = {https://mlanthology.org/ecmlpkdd/2013/azar2013ecmlpkdd-regret/}
}