{"title":"An Approach to the Development of a Game Agent Based on SOM and Reinforcement Learning","authors":"Keiji Kamei, Yuuki Kakizoe","doi":"10.1109/IIAI-AAI.2016.115","DOIUrl":null,"url":null,"abstract":"Recently, the researches that create agents which play board games have been studied actively. According to those studies, those agents have abilities that are comparable to the strongest experts. However, it can be said that those agents depend on the computational capability because that abilities of those agents are realized by thousands of lookahead search. On theotherhand, humanbeingshavenoadvantagescomparedwith numerical capability of computers, however, experts sometimes defeat those agents. In contrast to other approaches, our purpose is to create the agent which requires only low computational capability but is strong, like human beings. To realize our aim, we have proposed to develop the agent based on Self-Organizing Maps and reinforcement learning. From the experimental results, the agent learned by MC-learning achieved a 58% winning rate against the adversary program, so that we have succeeded in improving the winning rate over 10%.","PeriodicalId":272739,"journal":{"name":"2016 5th IIAI International Congress on Advanced Applied Informatics (IIAI-AAI)","volume":"52 1","pages":"0"},"PeriodicalIF":0.0000,"publicationDate":"2016-07-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"1","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"2016 5th IIAI International Congress on Advanced Applied Informatics (IIAI-AAI)","FirstCategoryId":"1085","ListUrlMain":"https://doi.org/10.1109/IIAI-AAI.2016.115","RegionNum":0,"RegionCategory":null,"ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"","JCRName":"","Score":null,"Total":0}
引用次数: 1
Abstract
Recently, the researches that create agents which play board games have been studied actively. According to those studies, those agents have abilities that are comparable to the strongest experts. However, it can be said that those agents depend on the computational capability because that abilities of those agents are realized by thousands of lookahead search. On theotherhand, humanbeingshavenoadvantagescomparedwith numerical capability of computers, however, experts sometimes defeat those agents. In contrast to other approaches, our purpose is to create the agent which requires only low computational capability but is strong, like human beings. To realize our aim, we have proposed to develop the agent based on Self-Organizing Maps and reinforcement learning. From the experimental results, the agent learned by MC-learning achieved a 58% winning rate against the adversary program, so that we have succeeded in improving the winning rate over 10%.