Intelligence Gaming Research Based on MCTS-RAVE
Yili Wang · Jiangnan daxue xuebao. Ziran kexue ban · 2011
For Monte-Carlo Tree Search(MCTS) algorithm,its intelligence heavily depends on the times of Monte-Carlo(MC) simulations and huge simulation times are needed to gain high intelligence.In this paper,a method of improving MCTS is proposed.Online reinforcement learning(RL) is added to MCTS.RL knowledge is accumulated through MC simulation,and Rapid Action Value Estimation(RAVE) is made.A case is presented to show that the MCTS-RAVE is more intelligent than MCTS.