Dear students,
Welcome to the Reinforcement Learning course.
To qualify for the final exam, you must complete three sets of exercises. The first sheet will be released in the second week of the semester, with the following two released in the second and final thirds of the semester. We will update this page with additional information upon their release and notify you via Discord. You must meet the minimum required points for each set.
Each exercise sheet will consist of two parts: a practical part, where you will implement the algorithms covered in the lectures, and an analytical part, where you will discuss the results of your experiments in the form of a report. The topics for the three sheets are:
- Value iteration and tabular methods (Due Date: November 4, 2024)
- DQN and its variants (Due Date: November 25, 2024)
- Policy gradient methods (Due Date: December 16, 2024)
You can work alone or in pairs (see details in the assignment).
The exercises are designed to align closely with both the lectures and the exams, so we strongly encourage you to approach them carefully. While passing the exercises above the minimum threshold can positively influence your final mark, it is still possible to achieve an A by meeting the threshold alone.
During the exercises, you will regularly work with Python libraries such as Gymnasium, NumPy, and PyTorch. Some example code will be provided in the exercise assignments, but we also recommend the PyTorch tutorial from UC Berkeley as additional material.
If you have any questions, feel free to reach out through the Discord server!