强化学习读书笔记 - 14 - 心理学

强化学习读书笔记 - 14 - 心理学

学习笔记:
Reinforcement Learning: An Introduction, Richard S. Sutton and Andrew G. Barto c 2014, 2015, 2016

参照

停在这里了

从这一章开始叫做Looking Deeper。
讲的有心理学(Psychology),神经科学(Neuroscience) 和强化学习的联系,
还有强化学习的应用和案例(Applications and case studies)和前沿(Frontiers)。

基本上需要大量的翻译。这不是我的特长。
所以我的笔记先停在这里了。

心理学(Psychology)

术语

  • reinforcement
    在心理学中,指动物接收到一个刺激(或者经历一个刺激的消失),行为和另一个刺激(或者反应)的关联模式得到了(强度或者频率上的)加强。
  • reinforcer - 强化刺激
  • reward - 奖赏
    让动物认知好行为的事物或者事件。
  • penalty - 惩罚
    让动物认知坏行为的事物或者事件。
  • reinforcement signal - 加强信号
    加强信号的一个例子:TD error。
  • action
  • control
    在强化学习中,控制是指本体影响它的环境,带来期望的状态或者事件。
  • stimulus-response learning - 刺激-反应学习
  • prediction algorithm
  • control algorithm
    Policy improvement algorithms
  • unconditioned responses
  • unconditioned stimulus
  • conditioned responses
  • conditioned stimulus
  • classical conditioning - 条件反射

算法列表

2
A simple bandit algorithm
4
Iterative policy evaluation
Policy iteration (using iterative policy evaluation)
Value iteration
5
First-visit MC policy evaluation (returns V  v)
Monte Carlo ES (Exploring Starts)
On-policy rst-visit MC control (for "-soft policies)
Incremental o -policy every-visit MC policy evaluation
O -policy every-visit MC control (returns   )
6
Tabular TD(0) for estimating v
Sarsa: An on-policy TD control algorithm
Q-learning: An o -policy TD control algorithm
Double Q-learning
7
n-step TD for estimating V  v
n-step Sarsa for estimating Q  q, or Q  q for a given 
O -policy n-step Sarsa for estimating Q  q, or Q  q for a given 
n-step Tree Backup for estimating Q  q, or Q  q for a given 
O -policy n-step Q() for estimating Q  q, or Q  q for a given 
8
Random-sample one-step tabular Q-planning
Tabular Dyna-Q
Prioritized sweeping for a deterministic environment
9
Gradient Monte Carlo Algorithm for Approximating ^v  v
Semi-gradient TD(0) for estimating ^v  v
n-step semi-gradient TD for estimating ^v  v
LSTD for estimating ^v  v (O(n2) version)
10
Episodic Semi-gradient Sarsa for Control
Episodic semi-gradient n-step Sarsa for estimating ^q  q, or ^q  q
Di erential Semi-gradient Sarsa for Control
Di erential semi-gradient n-step Sarsa for estimating ^q  q, or ^q  q
12
Semi-gradient TD() for estimating ^v  v
True Online TD() for estimating >  v
13
REINFORCE, A Monte-Carlo Policy-Gradient Method (episodic)
REINFORCE with Baseline (episodic)
One-step Actor-Critic (episodic)
Actor-Critic with Eligibility Traces (episodic)
Actor-Critic with Eligibility Traces (continuing)

posted @ 2017-03-28 19:03  SNYang  阅读(1746)  评论(0编辑  收藏  举报