Analysing And Improving The Knowledge-based Fast Evolutionary MCTS Algorithm
Analysing And Improving The Knowledge-based Fast Evolutionary MCTS Algorithm
复制标题
基于知识的快速进化MCTS算法分析与改进
DOI：
--
复制
发表时间：
2015
期刊：
影响因子：
--
通讯作者：
D. Thierens
中科院分区：
文献类型：
--
作者：
J. V. Eeden;D. Thierens
MSc.技术人工智能科学系信息和计算科学系Jim货车Eeden，3344800在不断努力达到人类水平的智能与智能代理，它已经变得很清楚，代理应该能够处理不同的问题域主机。在一个特定的领域非常好，比如国际象棋，是不够的。这就是为什么引入了通用视频游戏（GVGP）领域。在这里，代理将玩各种不同的视频游戏，而他们事先不知道他们将玩哪种游戏。代理人被迫通过玩来学习游戏。本硕士论文将研究为GVGP创建的蒙特卡罗树搜索算法的一个特定版本，称为基于知识的快速进化（KB Fast-Evo）MCTS算法。该算法使用一个知识库来存储有关游戏的信息，也是一个进化的方法来偏置游戏模拟。首先对MCTS算法进行了一般性的研究，并对一般视频游戏进行了详细的解释。然后对KB Fast-Evo算法进行了分析，发现该算法还有很大的改进空间。在最后几章中，我们尝试通过改变进化方法、增加路径查找和其他一些微调来改进KB Fast-Evo算法。最终算法的性能确实好了很多，但仍有改进的空间。然而，算法本身不太可能是最优的。
MSc. Technical Artificial Intelligence Faculty of Science Department of Information and Computing Science by Jim van Eeden, 3344800 In the ongoing strive to reach human-level intelligence with intelligent agents, it has become clear that the agents should be able to deal with a host of different problem domains. Being very good on one particular domain, like chess for instance, does not suffice. That is why the area of General Video Game Playing (GVGP) was introduced. Here, agents will play a wide range of different video games, while they do not know which game they will be playing beforehand. The agents are forced to learn the game by playing. This master thesis will research one particular version of the Monte Carlo Tree Search algorithm that was created for GVGP, called the Knowledge-based Fast Evolutionary (KB Fast-Evo) MCTS algorithm. This algorithm uses a knowledge base to store information about the game and also an evolutionary approach to bias game simulations. First a general study of the MCTS algorithm is given, together with a detailed explanation of General Video Game Playing. Then the KB Fast-Evo algorithm is analyzed, where a lot of room for improvement becomes apparent. In the final chapters an attempt is made to improve the KB Fast-Evo algorithm, by changing the evolutionary approach, by adding path finding and some other fine-tuning. The final algorithm does perform a lot better, but there is still room for improvement. However, it is not likely the algorithm will ever be optimal on its own.
适用于一般视频游戏的基于知识的快速进化 MCTS
DOI：
10.1109/cig.2014.6932868
发表时间：
2014
期刊：
--
影响因子：
--
作者：
Perez D
通讯作者：
Perez D