Acquisition of go knowledge from game records

  • Authors:

  • Takuya Kojima;Atsushi Yoshikawa

  • Affiliations:

  • NTT Communication Science Labs, Media Information Lab, 3-1 Morinosato Wakamiya, Atsugi, Kanagawa 243-0198, Japan;NTT data, Research and Development Headquarters, 1-21-2 Shinkawa, Chuo-Ku, Tokyo, 104-0033, Japan

  • Venue:
  • Machines that learn to play games
  • Year:
  • 2001

Quantified Score

Hi-index 0.00

Visualization

Abstract

A large amount of knowledge is considered very useful for systems playing such games as Go and shogi (Japanese chess) which have a much larger search space than chess. Knowledge acquisition is therefore necessary for systems playing these games. This paper explains two approaches to acquire Go knowledge from game records: a deductive approach and an evolutionary one. The former is taken to acquire strict knowledge; several rules are acquired from a single training example. The latter is taken to acquire a large amount of heuristic knowledge from a large amount of training examples. Tsume-go problems are solved in order to compare the performance of the algorithm with others.