Data Optimization Of Go Strategy Network Based On Machine Learning

There are thousands of years of history since Go has been created,The rules of this game is simple,but the difficulty is very high,computer Go decision-making has always been a major problem in the field of artificial intelligence,the game tree of Go has complexity of about 10 of the 300 power side.Only by ordinary algorithms and hard programming is unable to solve this problem.After adding a lot of Go knowledges,the traditional AI of go can not win the professional players.After team "DeepMind" applied deep convolution neural network in computer Go,the development of Go AI has made a significant leap.Now,improved Go AI can win the professional player.With the significant breakthrough in artificial intelligence in the computer Go,the convolution neural network model which DeepMind used is used to all walks of life.First of all,This paper introduces the basic principle of convolution neural network and how to apply convolution neural network technology to computer Go decision problem.,and provides the model of the policy network.Secondly,it describes what input is required for a convolutional neural network applied to a computer Go,introducing the whole process of obtaining the data set from the original game data.Finally,it tells the machine learning platform used to train the convolutional neural network designed in this paper and the way in which training is done and the problems encountered during training and how to overcome these problems.It describes and analyzed the results of the convolution neural network,the improved methods and measures are put forward to make the results better.