概要:

In this study, we evaluate the board in Othello game by using a perceptron which is a kind of neural network. Board evaluation is to express the advantage of board by the numerical value.

The network learns by back-propagation using about 50,000 game records that existing Othello programs fought mutually.

By changing the number of units in the layer, the network learn under several conditions to investigate the change in the learning efficiency.

It was found that the result is better with more units in the hidden layer, but the required number of learning times is increased.

To reduce the error of the last stage and to find a set of parameters to obtain a better result is for further study.