The game of Go is more challenging than other board games, due to the
difficulty of constructing a position or move evaluation function. In this
paper we investigate whether deep convolutional networks can be used to
directly represent and learn this knowledge. We train a large 12-layer
convolutional neural network by supervised learning from a database of human
professional games. The network correctly predicts the expert move in 55% of
positions, equalling the accuracy of a 6 dan human player. When the trained
convolutional network was used directly to play games of Go, without any
search, it beat the traditional search program GnuGo in 97% of games, and
matched the performance of a state-of-the-art Monte-Carlo tree search that
simulates a million positions per move.
Description
[1412.6564] Move Evaluation in Go Using Deep Convolutional Neural Networks
%0 Generic
%1 maddison2014evaluation
%A Maddison, Chris J.
%A Huang, Aja
%A Sutskever, Ilya
%A Silver, David
%D 2014
%K 2014 arxiv cnn deep-learning evolutionary
%T Move Evaluation in Go Using Deep Convolutional Neural Networks
%U http://arxiv.org/abs/1412.6564
%X The game of Go is more challenging than other board games, due to the
difficulty of constructing a position or move evaluation function. In this
paper we investigate whether deep convolutional networks can be used to
directly represent and learn this knowledge. We train a large 12-layer
convolutional neural network by supervised learning from a database of human
professional games. The network correctly predicts the expert move in 55% of
positions, equalling the accuracy of a 6 dan human player. When the trained
convolutional network was used directly to play games of Go, without any
search, it beat the traditional search program GnuGo in 97% of games, and
matched the performance of a state-of-the-art Monte-Carlo tree search that
simulates a million positions per move.
@misc{maddison2014evaluation,
abstract = {The game of Go is more challenging than other board games, due to the
difficulty of constructing a position or move evaluation function. In this
paper we investigate whether deep convolutional networks can be used to
directly represent and learn this knowledge. We train a large 12-layer
convolutional neural network by supervised learning from a database of human
professional games. The network correctly predicts the expert move in 55% of
positions, equalling the accuracy of a 6 dan human player. When the trained
convolutional network was used directly to play games of Go, without any
search, it beat the traditional search program GnuGo in 97% of games, and
matched the performance of a state-of-the-art Monte-Carlo tree search that
simulates a million positions per move.},
added-at = {2018-05-01T09:37:11.000+0200},
author = {Maddison, Chris J. and Huang, Aja and Sutskever, Ilya and Silver, David},
biburl = {https://www.bibsonomy.org/bibtex/21e2eb596d61001ace93bd2fc9ee3e6c1/achakraborty},
description = {[1412.6564] Move Evaluation in Go Using Deep Convolutional Neural Networks},
interhash = {545a33da64011919396aec8759682e80},
intrahash = {1e2eb596d61001ace93bd2fc9ee3e6c1},
keywords = {2014 arxiv cnn deep-learning evolutionary},
note = {cite arxiv:1412.6564Comment: Minor edits and included captures in Figure 2},
timestamp = {2018-05-01T09:37:11.000+0200},
title = {Move Evaluation in Go Using Deep Convolutional Neural Networks},
url = {http://arxiv.org/abs/1412.6564},
year = 2014
}