Reinforcement Learning (DS-RL-M)
Reinforcement learning (RL) is the field that enables intelligent agents to learn by doing: they interact with an environment, receive feedback in the form of rewards or penalties, and adapt their behaviour to maximise long-term success. From game-playing AIs like AlphaGo to autonomous robots navigating complex terrains, RL has become a cornerstone of modern artificial intelligence.
This introductory course is concerned with the core concepts, mathematical foundations, and practical algorithms that power reinforcement learning. Students will learn how to model sequential decision problems as Markov Decision Processes (MDPs), derive optimal policies using dynamic programming, and extend these ideas into scalable, data-driven solutions such as Monte-Carlo methods, Temporal-Difference learning, and policy gradients.
The lecture is accompanied by practicals and assignments that will help participants to understand the math behind reinforcement learning and gain practical, hands-on experience.
The following is a selection of topics that will be addressed in the course:
- Markov Decision Processes
- Dynamic Programming
- Multi-Armed Bandits and Exploration
- Monte Carlo Methods
- Temporal Difference Learning
- Planning and Model-based Reinforcement Learning
- Prediction with Function Approximation
- Control with Function Approximation
- Policy Gradient Methods
- Advanced Topics of Reinforcement Learning
Organisation
The lecture is accompanied by practicals and assignments and runs each winter term.
The course language is English.
Prerequisites: good working knowledge of programming (e.g., in Python)
Recommended (not mandatory) completion of modules: Einführung in die Künstliche Intelligenz/Introduction to Artificial Intelligence [KogSys-KI-B], Einführung in Maschinelles Lernen/Introduction to Machine Learning [KogSys-ML-B], Mathematics for Machine Learning [xAI-MML-M], Deep Learning [xAI-DL-M]
Degree Programmes
This module is available in the following degree programmes: