概要:

The purpose of this study is to create a convolution neural network using TensorFlow, which is a machine learning library developed by Google, and to conduct a deep learning to evaluate the board of the Othello game.

The game records of the Othello game fought by Othello AI's each other are entered into a perceptron and a convolution neural network created using TensorFlow and supervised learning is carried out with these records.

The errors of perceptron and convolution neural network are compared. After that, in order to reduce the error of the convolution neural network, how much error reduction can be achieved by setting dropout is examined.

Although the convolution neural network has less error than the perceptron, but it turned out that the processing itself takes a lot of time. Dropout can be said to be an effective means to reduce the error without increasing processing time there.

Discovery of parameter settings to obtain better results, and error reduction in the last stage are future tasks.