论文Control of Memory, Active Perception, and Action in Minecraft发表于ICML2016,该论文主要利用外部记忆结构(类似NTM,DNC,MemNN),通过将记忆存入外部记忆结构,提高Agent的迁移能力,导航能力可以泛化到unseen environment. 作者是密歇根大学安娜堡的博士生,他的另一篇文章《zero-shot task generalization with multi-task deep reinforcement learning》发表于ICML2017, 也是研究Agent在游戏环节下的迁移,大家有兴趣可以参看。

读论文-Control of Memory, Active Perception, and Action in Minecraft

读论文-Control of Memory, Active Perception, and Action in Minecraft

读论文-Control of Memory, Active Perception, and Action in Minecraft

读论文-Control of Memory, Active Perception, and Action in Minecraft

读论文-Control of Memory, Active Perception, and Action in Minecraft

读论文-Control of Memory, Active Perception, and Action in Minecraft

读论文-Control of Memory, Active Perception, and Action in Minecraft

读论文-Control of Memory, Active Perception, and Action in Minecraft

读论文-Control of Memory, Active Perception, and Action in Minecraft

读论文-Control of Memory, Active Perception, and Action in Minecraft

读论文-Control of Memory, Active Perception, and Action in Minecraft

读论文-Control of Memory, Active Perception, and Action in Minecraft

读论文-Control of Memory, Active Perception, and Action in Minecraft

读论文-Control of Memory, Active Perception, and Action in Minecraft

读论文-Control of Memory, Active Perception, and Action in Minecraft

读论文-Control of Memory, Active Perception, and Action in Minecraft

这张ppt显示的是FRMQN算法在搜索记忆的过程,每幅图中,上面的行是每帧游戏画面。

大家也可以参考这篇,写的很通俗易懂,而且十分有趣! https://zhuanlan.zhihu.com/p/21320865?refer=intelligentunit

相关文章:

  • 2021-10-13
  • 2021-10-31
  • 2021-12-27
  • 2021-04-04
  • 2021-05-31
  • 2021-10-23
  • 2021-05-20
  • 2021-07-06
猜你喜欢
  • 2021-04-25
  • 2021-08-16
  • 2021-08-08
  • 2021-11-02
  • 2021-11-30
  • 2021-12-20
  • 2021-08-17
相关资源
相似解决方案