Study On Several Problems For Computer Go System

Computer Game is an important branch of Artificial Intelligence research. Many methods and theories originate from it. It also brings remarkable social and scientific influence. After Computer Chess, Computer Go becomes the most active part of Computer Game research area recently.After summarizing related researches on Computer Go, we make thorough study on some key problems in this dissertation. The innovative points are as follows:1,Influence model is constructed to scale the influence that stones make, to estimate the areas of two players, and to divide stones into groups. According to the game situation evaluation based on influence model, the system can choose the key position of attack or defense. Three kinds of influence model are proposed in this dissertation, for refined calculation, static evaluation and searching respectively. It is an innovation to use different models in different modules in Computer Go system, and achieve good results in experiments.2,A machine learning method for extracting Go joseki for Computer Go system is proposed in this dissertation. By using this method, various go joseki can be automatically extracted from human-played Go game files so that we can construct a joseki-tree database. 34,000 human-played Go game files are processed and 680,638 joseki nodes are created. We integrate the information of joseki recorded in learning process and static evaluation function into a new proper joseki choosing method. This method connects joseki to game situation tightly, and makes better use of joseki in practice.3,For move selection, static evaluation function, which is based on influence model, is taken as the core method. We also make use of searching module to find the status of groups of stones, and take joseki information into consideration. In this way, our system can simulate the intuition of human players, and take advantages of the computing ability of computer. The system has some level of intelligence bycombining static analysis, dynamical searching and knowledge database application together.Based on the above work, we design and implement an integrated Computer Go system. The system reaches world level on some aspects, such as joseki selection, life-and-death problems, etc.