Reinforcement Learning
Welcome to the last lecture: Lecture 6 – Reinforcement Learning. Here we will use what we covered about MDPs last lecture and answer the question what happens if our agent doesn’t know the transition functions or the reward functions. The agent must explore and learn. We will cover passive reinforcement learning, where the policy of the agent is fixed while the agent learns transitions and rewards or values and q-values directly, and active reinforcement learning where the agent must additionally also decide which actions to take and thereby balance exploration of the (yet) unknown and exploitation of the (currently) best option. In a last step we will also cover feature based Reinforcement Learning to make it scalable for larger problems.
Topics in this course:
- Introductory Examples (RL 1)
- RL Taxonomy (RL 2)
- Model-Based RL (RL 3)
- Model-Free RL
- Direct Evaluation (RL 4)
- TD Learning (RL 5)
- Q-Learning (RL 6)
- Active RL (RL 7)
- Approximate Q-Learning (RL 8, 9)
- Policy Search (RL 10)
The lecture videos are available here.
Step-by-step examples and additional explanations:
- TD Learning
- Q Learning
- Approximate Q-Learning