Research On Machine Learning And The Application In Computer Go

Artificial intelligence has already been fully applied to various fields in the modernsociety, such as aerospace, military, manufacturing, and commercial, medical and so on.While the initially target of machine learning is to let the computer system have theperson’s ability of learning, then the computer system achieve the artificial intelligence. Asbeing the carrier of studying artificial intelligence, computer game has a good advantage asthe fly’s advantage in the study of genetic gene. To this end, computer game is also knownas the ’ fly ’ in artificial intelligence research. As many games in one, Go has a complicatedstate space. This directly means that the research of the computer Go has considerabledifficulty. Introduce machine learning algorithm to the computer game can let the computerGo system have self learning ability, so as to achieve the intelligent game.The human learning is an important intelligent behavior, such as a child learning totalk, to walk and so on. H.A.Simon, who is the master of Artificial intelligence master,once pointed out that, learning is the system increase or improves itself ability in repetitivework. With the increasing development of computer technology, the capacity of humancollecting data, information has been greatly improved. How to learn the requiredknowledge with this information? Machine learning can give the answer. At present, themachine learning has been another important area in artificial intelligence applicationresearch after the expert system.This article look the computer go system as an agent, the agent can learn the availableknowledge through the record of playing chess (knowledge) and the simulating playingchess, then look for the optimal pace at the present or within several steps in order tovictory, and ultimately complete playing chess. This process we always conceptualized as asearch process. In this search process, agent can learn knowledge through a certain method,and finally obtain the available knowledge, and achieve the goal. This article mainlyanalyses the following problems:1. Introduce the game concept, the commonly used algorithms of computer game, as wellas the characteristics of computer go and it’s key technology. Analyze the importantrole of machine learning for intelligent, as well as the significance of introducing themachine learning algorithm to the computer go.2. Introduce the solution of using machine learning algorithm in computer go, such as Monte-Carlo, neural network, and the reinforcement learning. Mainly introduce themechanism, application of temporal difference algorithm, which is belonged tomachine learning area. Create a computer go system model based on the timedifference algorithm, and apply it to computer go;3. Improve the temporal difference algorithm which is applied in computer go. Take thechessboard state value which has already learned as one of the rewards after choosingaction. This makes the agent get more truth board information.4. Adopt ongoing training to improve game capability gradually. Through a large numberof experiments and training and the constantly optimizing the model parameters,improve the agent’s game level and strengthen the actual effect of the algorithm. Andprove the feasibility of the application and improvement of temporal difference incomputer go.