Deep Reinforcement Learning

Author

Julien Vitay

Overview

This website contains the materials for the module Deep Reinforcement Learning taught by Dr. Julien Vitay at the Technische Universität Chemnitz, Faculty of Computer Science, Professorship for Artificial Intelligence.

Each section/lecture is accompanied by a set of videos, the slides and a written version of the content. The videos are integrated in the lecture notes, but you can also access the complete playlist on Youtube.

Exercises are provided in the form of Jupyter notebooks, allowing to implement in Python at your own pace the algorithms seen in the lectures and to learn to use reinforcement learning libraries such as gym. A notebook to work on (locally or on Colab) and the solution are available in the Exercises section.

Syllabus

Lectures

  1. Introduction
    1. Introduction
    2. Math basics (optional)
  2. Tabular RL
    1. Bandits
    2. Markov Decision Processes
    3. Dynamic Programming
    4. Monte Carlo control
    5. Temporal Difference
    6. Function approximation
    7. Deep learning
  3. Model-free RL
    1. Deep Q-network
    2. Beyond DQN
    3. Policy Gradient
    4. A2C / A3C
    5. DDPG
    6. TRPO / PPO
    7. SAC
  4. Model-based RL
    1. Model-based RL
    2. Learned models
    3. AlphaGo
    4. Successor representations (optional)
  5. Outlook
    1. Outlook

WS2023-24: Here are the compressed slides for A3C/DDPG/PPO: html. SAC is unfortunately skipped.

Exercises

Notebooks and videos are in the List of Exercises. Below are links to the rendered solutions.