A (Long) Peek into Reinforcement Learning
Lilian Weng
This article provides an educational overview of reinforcement learning fundamentals, covering key concepts including agents, environments, policies, value functions, and Markov decision processes. The article explains that RL agents learn optimal strategies by interacting with environments to maximize cumulative rewards, using examples like AlphaGo defeating professional Go players and OpenAI's bot winning DOTA2 matches. The material introduces formal mathematical frameworks including Bellman equations for computing value functions, distinguishing between model-based and model-free approaches, and categorizing algorithms by whether they rely on known or learned environment models.
Why it matters
[Updated on 2020-09-03: Updated the algorithm of SARSA and Q-learning so that the difference is more pronounced. [Updated on 2021-09-19: Thanks to 爱吃猫的鱼, we have this post in Chinese].
