Reinforcement learning: a comparison of UCB versus alternative adaptive policies

DOI10.1515/9783110663075-006MaRDI QIDQ3300065zbMATH OpenOpenAlexFDO

Authors Wesley Cowan, Michael N. Katehakis, Daniel Pirutinsky

Publication date 27 July 2020

Published in First Congress of Greek Mathematicians (Search for Journal in Brave)

Full work available at URL https://arxiv.org/abs/1909.06019

zbMATH Keywords

unknown transition probabilities. deterministic minimum empirical divergence, posterior sampling

Mathematics Subject Classification ID

Markov and semi-Markov decision processes (90C40)

Abstract: In this paper we consider the basic version of Reinforcement Learning (RL) that involves computing optimal data driven (adaptive) policies for Markovian decision process with unknown transition probabilities. We provide a brief survey of the state of the art of the area and we compare the performance of the classic UCB policy of cc{bkmdp97} with a new policy developed herein which we call MDP-Deterministic Minimum Empirical Divergence (MDP-DMED), and a method based on Posterior sampling (MDP-PS).

Recommendations

• An asymptotically optimal policy for finite support models in the multiarmed bandit problem
• Finite-time analysis of the multiarmed bandit problem
• Bayesian Reinforcement Learning with Exploration
• Reinforcement learning in finite MDPs: PAC analysis
• Reinforcement learning: a tutorial survey and recent advances

Cited in

(2)

• Sequential decision-making under uncertainty: a robust MDPs review
• Complexity bounds for deterministic partially observed Markov decision processes

This page was built for publication: Reinforcement learning: a comparison of UCB versus alternative adaptive policies

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q3300065)