Deep Reinforcement Learning from Self-Play in Imperfect-Information Games

Heinrich, Johannes, and David Silver. "Deep reinforcement learning from self-play in imperfect-information games." arXiv preprint arXiv:1603.01121(2016).

这篇文章提出了基于深度学习的自我博弈达到纳什均衡的训练方法。这个方法避免了人为的先验知识的误导,采用了端到端的训练方式,达到了人类专家级水平。

方法:

通过自我博弈产生训练数据,用来训练Qlearning网络和有监督学习网络。然后对这两个网络做ensemble

原文地址:https://www.cnblogs.com/huangshiyu13/p/7347968.html