Learning finite functions by neural networks : Evaluation of Pentago positions by convolutional neural networks (Algebras, logics, languages and related areas)

Abstract

A convolution neural network (CNN) is a useful tool that approximates a finite function. It is used as a solver for various problems in the real world. In this paper, results of experiments on training variations of a small CNN used for image recognition for evaluating Pentago positions are mainly reported. The author hopes that the results are used in discussion of applicability of deep neural networks to researches in theoretical computer science.