Confidence Backup Updates for Aggregating MDP State Values in Monte-Carlo Tree Search

Zahy Bnaya, Alon Palombo, Rami Puzis, Ariel Felner

Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review

Abstract

Monte-Carlo Tree Search (MCTS) algorithms estimate the value of MDP states based on rewards received by performing multiple random simulations. MCTS algorithms can use different strategies to aggregate these rewards and provide an estimation for the states’ values. The most common aggregation method is to store the mean reward of all simulations. Another common approach stores the best observed reward from each state. Both of these methods have complementary benefits and drawbacks. In this paper, we show that both of these methods are biased estimators for the real expected value of MDP states. We propose an hybrid approach that uses the best reward for states with low noise, and otherwise uses the mean. Experimental results on the Sailing MDP domain show that our method has a considerable advantage when the rewards are drawn from a noisy distribution.

Original languageEnglish
Title of host publicationProceedings of the 8th Annual Symposium on Combinatorial Search, SoCS 2015
EditorsLevi Lelis, Roni Stern
PublisherAAAI press
Pages156-160
Number of pages5
ISBN (Electronic)9781577357322
StatePublished - 1 Jan 2015
Event8th Annual Symposium on Combinatorial Search, SoCS 2015 - Ein Gedi, Israel
Duration: 11 Jun 201513 Jun 2015

Publication series

NameProceedings of the 8th Annual Symposium on Combinatorial Search, SoCS 2015
Volume2015-January

Conference

Conference8th Annual Symposium on Combinatorial Search, SoCS 2015
Country/TerritoryIsrael
CityEin Gedi
Period11/06/1513/06/15

ASJC Scopus subject areas

  • Computer Networks and Communications

Fingerprint

Dive into the research topics of 'Confidence Backup Updates for Aggregating MDP State Values in Monte-Carlo Tree Search'. Together they form a unique fingerprint.

Cite this