更改

增强学习-入门导读

添加3,091字节2019年5月23日 (四) 07:56
== 教材 =强化学习 =
# Richard S. Sutton, Andrew Barto, An Introduction to Reinforcement Learning, MIT Press, 1998. [http://webdocs.cs.ualberta.ca/~sutton/book/the-book.html Intro_RL]== 强化学习定义 ==# Csaba Szepesvari, Algorithms for Reinforcement Learning, Synthesis lectures on artificial intelligence and machine learning 4, no. 1, pp.1强化学习(Reinforcement Learning)是一种通用的决策框架( decision-103, 2010. [http://www.ualberta.ca/~szepesva/papers/RLAlgsInMDPs.pdf RLAlgsInMDPs]making framework)。
== 研究 ==Agent代理具有采取动作(action)的能力(capacity),每次动作都会影响Agent的未来状态(State),返回一个标量的奖赏信号(reward signal)来量化表示成功与否(success)。
强化学习算法的目标(Goal)就是如何采取动作(action)最大化未来的奖赏(future reward)。 == 强化学习要素 == 从强化学习Agent的角度看,强化学习包含一组组件: (1) 策略(Policy)是指:Agent的行为函数; (2) 价值函数(Value)是指:每个状态与动作的成效如何? (3) 模型(Model): Agent的环境的表示。 == 通用人工智能AGI == 深度强化学习(Deep Reinforcement Learning, Deep RL)就是把强化学习RL和深度学习DL的结合起来。 用强化学习定义目标,用深度学习给出相应的机制,如Q学习等技术,以实现通用人工智能(Artificial General Intelligence, AGI)。 = 强化学习应用 = == 计算机围棋 AlphaGo == # Mastering the game of Go with deep neural networks and tree search, nature 2015.# Better Computer Go Player with Neural Network and Long-term Prediction, ICLR 2016.# Pachi: State of the art open source Go program, Advances in computer games, Springer Berlin Heidelberg, 2011. ===多臂赌博机===
* 多臂赌博机(mutiarmed bandit problem)
#Multi-armed bandits with episode context, AMAI 2011.#Algorithms for Infinitely Many-Armed Bandits, nips 2009.一个赌徒面前有一系列老虎机(或赌博机),每个赌博机在投入硬币后,返回的回报是不同的。赌徒面临的问题是如何最大化自己的收益。
当赌徒尝试了一系列赌博机后,会获得一些统计上的收益。但是,赌徒并不知道赌博机背后的真实收益分布。在获得已有的收益后,赌徒遇到的策略问题是:是继续专注于当前获得的收益呢,还是去尝试更多的赌博机?
 
赌徒如果专注于已获得收益的赌博机,至少可以保持一定的收益。如果去尝试更多的先前未测试的赌博机,有可能出现尝试失败的情况,但也有可能会发现具有更大收益的赌博机。
 
UCB方法是针对多臂赌博机问题的一种解法,力图在在探索(在未知的赌博机)和遵从(现有经验)之间找到平衡。UCB 方法全称是(“Upper Confidence Bounds”), 即上置信边界方法。
 
'''UCB算法最早由以下论文提出。'''
 
#Peter Auer, Nicolo Cesa-Bianchi, and Paul Fischer. "Finite-time analysis of the multiarmed bandit problem." Machine learning 47, no. 2-3: 235-256, 2002.
 
'''PUCT算法由以下论文研究提出。'''
#Christopher D. Rosin, Multi-armed bandits with episode context, Annals of Mathematics and Artificial Intelligence, March 2011, Volume 61, Issue 3, pp 203–230 2011.
 
'''其它论文'''
#Wang, Yizao, Jean-Yves Audibert, and Rémi Munos. "Algorithms for infinitely many-armed bandits." Advances in Neural Information Processing Systems. 2009.
 
===蒙特卡洛树搜索===
* 蒙特卡洛树搜索(Monte-Carlo Tree Search)
# '''Monte-Carlo tree search and rapid action value estimation in computer Go, Artificial Intelligence, Elsevier 2011.'''
===卷积网络下围棋===* 神经网络卷积网络
# Mimicking Go Experts with Convolutional Neural Networks, ICANN 2008.
# '''Training Deep Convolutional Neural Networks to Play Go, ICML 2015.'''
* 进展在用3千万5dan以上的选手的棋局训练卷积网路,其中机器也会把人类选手下的昏招或者臭招也学会了。但是可以用自我博弈出的棋局数据来训练,这样就可以稀释掉这些昏招。 == 历史性进展 ==
# Achieving Master Level Play in 9 × 9 Computer Go, AAAI 2008.
# The grand challenge of computer Go Monte Carlo tree search and extensions, CACM 2012.
 
=阿尔法围棋=
 
# '''Mastering the game of Go with deep neural networks and tree search, Nature 2016.'''
#AlphaGo Zero
#AlphaZero
===计算机游戏===
#Mnih, Volodymyr, Koray Kavukcuoglu, David Silver, Andrei A. Rusu, Joel Veness, Marc G. Bellemare, Alex Graves et al. "Human-level control through deep reinforcement learning." Nature 518, no. 7540 (2015): 529-533.
=参考资料 = = 神经科学 =参考教材 == # Richard S. Sutton, Andrew Barto, An Introduction to Reinforcement Learning, MIT Press, 1998. [http://webdocs.cs.ualberta.ca/~sutton/book/the-book.html Intro_RL]# Csaba Szepesvari, Algorithms for Reinforcement Learning, Synthesis lectures on artificial intelligence and machine learning 4, no. 1, pp.1-103, 2010. [http://www.ualberta.ca/~szepesva/papers/RLAlgsInMDPs.pdf RLAlgsInMDPs] == 参考课程 ==
#UC Berkeley CS 294: Deep Reinforcement Learning, [1] Gadagkar, Vhttp://rll., Puzerey, P., Chen, R., Baird-daniel, E., Farhang, A., & Goldberg, J. (2016). Dopamine Neurons Encode Performance Error in Singing Birds. Science, 354(6317), 1278–1282berkeley.edu/deeprlcourse/ Deep RL]
行政员管理员
6,105
个编辑