Multilayer Stochastic Weight Assignment Network And Its Application To Chess Situation Judgement

Machine game is a hot and challenging research direction in the field of artificial intelligence?AI?,and machine learning is a commonly used research method.It is a hot research topic in machine game to apply machine learning methods to explore the unknown domain knowledge and has important theoretical and practical value.In this thesis we use multilayer stochastic weight assignment network?MSWAN?to study the issue of Chinese chess's situation judgment.MSWAN is a kind of multiple hidden-layer feedforward neural network which includes two components:auto encoder?AE?used as unsupervised feature extractor and supervised stochastic weight assignment network?SWAN?used as classifier.The main contributions of this paper include three aspects:?1?A feature extraction method for Chinese chess situation was proposed in this paper,and the proposed method has better representative ability.The Chinese chess's situation can be effectively judged by using the extracted features;?2?An approach to judge the endgame situation of Chinese chess by applying MSWAN was proposed and the proposed approach was experimentally compared with the simple SWAN on the accuracy of situation judgment.It is concluded that the MSWAN can effectively solve the problem of Chinese chess's situation judgment and its accuracy of situation judgment is higher than the one of simple SWAN;?3?An improved method for the feature extractor of MSWAN by L_1/2 regularization was proposed,and the proposed method was used to judge the Chinese chess's middlegame situation.Compared with the original model without L_1/2 regularization,the features extracted by the proposed method are more representative and the accuracy of Chinese chess's situation judgment is higher.