A Hybrid PAC Reinforcement Learning Algorithm
Machine Learning
2021-01-29 v2 Machine Learning
Abstract
This paper offers a new hybrid probably approximately correct (PAC) reinforcement learning (RL) algorithm for Markov decision processes (MDPs) that intelligently maintains favorable features of its parents. The designed algorithm, referred to as the Dyna-Delayed Q-learning (DDQ) algorithm, combines model-free and model-based learning approaches while outperforming both in most cases. The paper includes a PAC analysis of the DDQ algorithm and a derivation of its sample complexity. Numerical results are provided to support the claim regarding the new algorithm's sample efficiency compared to its parents as well as the best known model-free and model-based algorithms in application.
Cite
@article{arxiv.2009.02602,
title = {A Hybrid PAC Reinforcement Learning Algorithm},
author = {Ashkan Zehfroosh and Herbert G. Tanner},
journal= {arXiv preprint arXiv:2009.02602},
year = {2021}
}