Blogs of Reinforcement Learning

Contact me

• Blog -> https://cugtyt.github.io/blog/index
• Email -> cugtyt@qq.com
• GitHub -> Cugtyt@GitHub

---

本系列博客主要是关于强化学习笔记。

主要内容来源于：论文以及教程（Thomas Simonini Deep Reinforcement Learning Course with Tensorflow， Arthur Juliani Simple Reinforcement Learning with Tensorflow series），OpenAI Spinning Up in Deep RL

---

OpenAI Spinning Up Part 1： Key Concepts in RL

术语以及概念

• 状态和观察(states and observations)
• 行为空间(action spaces)，
• 策略(policies)，
• 轨迹(trajectories)，
• 不同的回报方式(different formulations of return)，
• RL优化问题(the RL optimization problem)，
• 值函数(value functions)。

---

Thomas Simonini Part 5: An intro to Advantage Actor Critic methods: let’s play Sonic the Hedgehog

The problem with Policy Gradients, How Actor Critic works, A2C and A3C, A2C in practice

---

Thomas Simonini Part 4: An introduction to Policy Gradients with Cartpole and Doom

Two types of policy, Advantages, Disadvantages, Policy Search, Monte Carlo Policy Gradients

---

Thomas Simonini Part 3+: Improvements in Deep Q Learning: Dueling Double DQN, Prioritized Experience Replay, and fixed Q-targets

Fixed Q-targets, Double DQNs, Dueling DQN, Prioritized Experience Replay (PER), Doom Deathmatch agent

---

Thomas Simonini Part 3: An introduction to Deep Q-Learning: let’s play Doom

Preprocessing part, The problem of temporal limitation, Experience Replay

---

Thomas Simonini Part 2: Diving deeper into Reinforcement Learning with Q-Learning

Q-learning algorithm: learning the Action Value Function, The Q-learning algorithm Process, Q* Learning with FrozenLake

---

Thomas Simonini Part 1: An introduction to Reinforcement Learning

Reinforcement Learning Process, Reward Hypothesis, Episodic or Continuing tasks, Monte Carlo vs TD Learning methods, Exploration/Exploitation trade off, Three approaches to Reinforcement Learning

---

Arthur Juliani Part 8 - Asynchronous Actor-Critic Agents (A3C)

The 3 As of A3C, Implementing the Algorithm

---

Arthur Juliani Part 7 - Action-Selection Strategies for Exploration

Greedy Approach, Random Approach, ϵ-Greedy Approach, Boltzmann Approach, Bayesian Approaches (w/ Dropout)

---

Arthur Juliani Part 6 - Partial Observability and Deep Recurrent Q-Networks

limited, changing world, Recurrent Neural Networks, Implementing in Tensorflow

---

Arthur Juliani Part 4 - Deep Q-Networks and Beyond

Double DQN, Dueling DQN

---

Arthur Juliani Part 3 - Model-Based RL

Model-Based

---

Arthur Juliani Part 2 - Policy-based Agents

Full reinforcement agent, Markov Decision Process, Cart-Pole Task

---

Arthur Juliani Part 1.5 - Contextual Bandits

Multi-armed bandit, Contextual bandit, Full RL problem, Contextual bandit代码

---

Arthur Juliani Part 1 - Two-armed Bandit

RL问题, Learning a Policy, Policy Gradients, Value functions, e-greedy policy, policy loss equation, The Multi-armed bandit 代码

---

Arthur Juliani Part 0 - Q-Learning Agents

Policy Gradient, Q-Learning, Bellman Equation, Q-Table 代码, Q-Learning NN 代码

---