Summary
International Symposium on Nonlinear Theory and Its Applications
2016
Session Number:C2L-E
Session:
Number:C2L-E-5
Efficient Board Feature Extraction for Strategy Improvement in Computer Go
Hayato Mitsuoka, Koujin Takeda,
pp.-
Publication Date:2016/11/27
Online ISSN:2188-5079
Summary:
Recently, significant progress was made in computer Go by deep learning. However, huge computer resource is required for achieving professional player's skill at present, which seems over-engineering for feature extraction on Go board. From this background, we discuss how to construct an efficient feature extraction method on Go board under deep learning framework. By making use of knowledge on image recognition by deep learning, we propose a method to reduce computational cost of board feature extraction without degrading Go playing performance.