RL【1】：Basic Concepts

系列文章目录

---

---

前言

本系列文章主要用于记录 B站 赵世钰老师的【强化学习的数学原理】的学习笔记，关于赵老师课程的具体内容，可以移步：
B站视频：【【强化学习的数学原理】课程：从零开始到透彻理解（完结）】
GitHub 课程资料：Book-Mathematical-Foundation-of-Reinforcement-Learning

---

Fundamental concepts in Reinforcement Learning

• State: The status of the agent with respect to the environment.
• State space: the set of all states. S={si}i=1NS = \{ s_i \}^N_{i=1}S={si​}i=1N​
• Action: For each state, an action refers to a set of possible moves or operations that the agent can take.
• Action space of a state: the set of all possible actions of a state. A(si)={ai}i=1NA(s_i) = \{a_i\}^N_{i=1}A(si​)={ai​}i=1N​
• State transition: When taking an action, the agent may move from one state to another. Such a process is called state transition. State transition defines the interaction with the environment.
• Policy $\pi$: Policy tells the agent what actions to take at a state.
• Reward: a real number we get after taking an action.
  • A positive reward represents encouragement to take such actions.
  • A negative reward represents punishment to take such actions.
  • Reward can be interpreted as a human-machine interface, with which we can guide the agent to behave as what we expect.
  • The reward depends on the state and action, but not the next state.
• Trajectory: A trajectory is a state-action-reward chain.
• Return: The return of this trajectory is the sum of all the rewards collected along the trajectory.
• Discount rate: The discount rate is a scalar factor $\gamma \in [0,1)$ that determines the present value of future rewards. It specifies how much importance the agent assigns to rewards received in the future compared to immediate rewards.
  • A smaller value of $\gamma$ makes the agent more short-sighted, emphasizing immediate returns.
  • A value closer to 1 encourages long-term planning by valuing distant rewards nearly as much as immediate ones.
• Discounted return $G_t$: The discounted return is the cumulative reward the agent aims to maximize, defined as the weighted sum of future rewards starting from time step $t$
  • $G_t=R_{t+1}+\gamma R_{t+2}+{\gamma }^2{R}_{t+3}+\cdots ={\sum }_{k=0}^{\infty }{\gamma }^k{R}_{t+k+1}.$
  • It captures both immediate and future rewards while incorporating the discount rate $\gamma$, thereby balancing short-term and long-term gains.
• Episode: When interacting with the environment following a policy, the agent may stop
  at some terminal states. The resulting trajectory is called an episode (or a
  trial).
  • An episode is usually assumed to be a finite trajectory. Tasks with episodes are called episodic tasks.
  • Some tasks may have no terminal states, meaning the interaction with the environment will never end. Such tasks are called continuing tasks.
  • In fact, we can treat episodic and continuing tasks in a unified mathematical way by converting episodic tasks to continuing tasks.
    • Option 1: Treat the target state as a special absorbing state. Once the agent reaches an absorbing state, it will never leave. The consequent rewards $r=0$.
    • Option 2: Treat the target state as a normal state with a policy. The agent can still leave the target state and gain $r=r+1$ when entering the target state.

Markov decision process (MDP)

• Sets:
  • State: the set of states $\mathcal{S}$
  • Action: the set of actions $\mathcal{A}(s)$ associated with state $s\in \mathcal{S}$
  • Reward: the set of rewards $\mathcal{R}(s,a)$
• Probability distribution:
  • State transition probability: at state $s$, taking action $a$, the probability of transitioning to state $s^{\prime }$ is $p(s^{\prime }|s,a)$
  • Reward probability: at state $s$, taking action $a$, the probability of receiving reward $r$ is $p(r|s,a)$
• Policy: at state $s$, the probability of choosing action $a$ is $\pi (a|s)$
• Markov property: memoryless property
  • $p(s_{t+1}|a_t,s_t,...,a_0,s_0)=p(s_{t+1}|a_t,s_t)$
  • $p(r_{t+1}|a_t,s_t,...,a_0,s_0)=p(r_{t+1}|a_t,s_t)$

---

总结

第一节系统性地介绍了 RL 中常见的一些概念，并做了通俗易懂的解释。同时，结合了马尔可夫决策过程，用数学公式化的语言进一步讲解了各概念的含义，为后续内容的讲解做好了铺垫。