wangshusen/drl

This repository is a course on deep reinforcement learning, organized as a series of lecture slides and accompanying video recordings. The videos are in Chinese, and the slides are PDF files available directly from the repository. It is structured as a university-style curriculum rather than runnable code. Reinforcement learning is a branch of machine learning where a software agent learns to make decisions by trying things out and receiving feedback, similar to how a person learns a game by playing it repeatedly. "Deep" reinforcement learning means the agent uses a neural network to process what it sees and decide what to do next, which allows it to handle far more complex situations than older rule-based approaches. The course moves through eight major topic areas. It opens with a conceptual overview of how reinforcement learning works, covering the main families of approaches: value-based methods, policy-based methods, and actor-critic methods. It also includes a session on AlphaGo to show how these ideas apply to a well-known real-world system. From there it goes into TD learning, which is a specific technique for estimating how good a given situation is by looking at outcomes a few steps ahead rather than waiting until the end of a game or task. This section covers Sarsa, Q-learning, and multi-step methods. Later sections go deeper into value-based approaches including experience replay and double DQN, then into policy gradient methods including REINFORCE and A2C. A section on continuous action spaces covers scenarios where the agent does not just pick from a fixed list of options but chooses values along a range. The final sections introduce multi-agent settings, where several agents interact, and imitation learning, where an agent learns by observing examples rather than by trial and error. This is a self-study or classroom resource. There is no software to install and no coding exercises included in the repository itself.