Analyses of Tabular AlphaZero on NoGo

Chu Hsuan Hsueh, Kokolo Ikeda, Sang Gyu Nam, I. Chen Wu

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

1 Scopus citations

Abstract

The AlphaZero algorithm has been shown to achieve superhuman levels of plays in chess, shogi, and Go. This paper presents analytic investigations of the algorithm on NoGo, a variant of Go that players cannot capture the opponents' stones. More specifically, lookup tables are employed for learning instead of deep neural networks, referred to as tabular AlphaZero. One goal of this work is to investigate how the algorithm is influenced by hyper-parameters. Another goal is to investigate whether the optimal plays and theoretical values can be learned. One of the hyper-parameters is thoroughly analyzed in the experiments. The results show that the tabular AlphaZero can learn the theoretical values and optimal plays in many settings of the hyper-parameter. Also, NoGo on different board sizes is compared, and the learning difficulty is shown to relate to the game complexity.

Original languageEnglish
Title of host publicationProceedings - 25th International Conference on Technologies and Applications of Artificial Intelligence, TAAI 2020
PublisherInstitute of Electrical and Electronics Engineers Inc.
Pages254-259
Number of pages6
ISBN (Electronic)9781665403801
DOIs
StatePublished - Dec 2020
Event25th International Conference on Technologies and Applications of Artificial Intelligence, TAAI 2020 - Taipei, Taiwan
Duration: 3 Dec 20205 Dec 2020

Publication series

NameProceedings - 25th International Conference on Technologies and Applications of Artificial Intelligence, TAAI 2020

Conference

Conference25th International Conference on Technologies and Applications of Artificial Intelligence, TAAI 2020
Country/TerritoryTaiwan
CityTaipei
Period3/12/205/12/20

Keywords

  • AlphaZero
  • NoGo
  • Tabular

Fingerprint

Dive into the research topics of 'Analyses of Tabular AlphaZero on NoGo'. Together they form a unique fingerprint.

Cite this