强化学习课件.pdf本课件讲解了强化学习的基本问题，经典Q学习理论，深度Q学习理论和程序讲解与训练。

文件名称: 强化学习课件.pdf

所属分类: 深度学习

开发工具:

文件大小: 2mb

下载次数: 0

上传时间: 2019-08-24

提供者: home****

下载 (2mb)

不能下载？报告错误

详细说明：本课件讲解了强化学习的基本问题，经典Q学习理论，深度Q学习理论和程序讲解与训练。强化学习相关参考资料网络资源 01 https://www.intelnervana.com/demystifying-deep-reinforcement-learning/ http://artint.info/html/artint265.html 参考文献 02 Playing Atari with Deep Reinforcement Learning 2013: arXiv: 1312.5602v1 Continuous control with deep reinforcement learning 2016: arXiv: 1509.02971v5 Human-level control through deep reinforcement learning 2015: Nature 14236 Mastering the game of Go without human knowledge 2017: Nature 24270 视频及网上课程 03 http://videolectures.net/rldm2015silverreinforcementlearning http://wwwo.cs.ucl.ac.uk/staff/d.silver/web/tEachinG.html http://rll.berkeley.edu/deeprlcourse/ 目录 contents01强化学习的基本问题 02经典Q学习理论 03深度Q学习理论 04程序讲解与训练 PART 01 强化学习的基本问题 Problems of Learning 强化学习的基本间题 https://www.technologyreview.com/s/603029/a-3-d-world-for-smarter-ai-agents/ .DeepMind Auxiliary Tasks State(s) ● Live Play 01 Reward 当前状态(图像、声音、向量等) 2+x Pixel Control Reward (r) 02 奖赏与激励 Action (a Reward Prediction Value Function Replay 03 动作(向前,向后,向左,向右) 100122 Value function(Q) Reward 04 评价函数Q(sa) No Reward Actions Value Function 强化学习的基本间题 Supervised Reinforcement UnSupervised Learning Learning Learnin g Dense Sparse NO Label Label Label Time-delayed Label( state i, Action k)=i good, bad] 监督还是非监督?是个问题强化学习的基本间题 thm algo o The agent has to exploit what it already E-greedy knows in order to obtain reward but it also 冒险-保守两难 has to explore in order to make better action Explore-Exploit Dilema selections in the future 0 o Dilemma neither exploitation nor exploration can be pursued exclusively 功劳归属问题 without failing at the task Credit Assignment Problem 02 A需要充分利用( explo已有知识去获得奖劢,但它也必须探索( explore)一些更佳的动作选项以便获得更好的奖赏两难:只坐享其成( exploi)或只冒险探索( explore)都可能导致任务失败两大问题强化学习的基本间题 thm algo Action/order 1st 2nd 3rd 4th 5th E-greedy a What chain Take a car of actions 冒险-保守两难 Take a bike resulted in Explore-Exploit Dilema Open door 0 reward Sit down ■ Which of the Shift in seat action to the Point at menu 功劳归属问题 right got you liter Credit Assignment Problem your steak? Tapped the table 02 Which of the preceding actions was responsible for getting the reward and to what extent? Value Function Q learning 哪个动作导致获得奖励,以何种程度? 十五的月亮,照在家乡照在边关有你的一半,也有我的一半。” 去两大问题 PART 02 经典Q学习理论 Classical Q-Learning

(系统自动生成,下载前可以参看下载内容)