Application will open in early 2026.
The application deadline will be April 24, 2026.
How do machines learn to make the right choices? How can individuals, firms, organizations, and researchers use automated decision-making?
This course provides an introduction and intuition for designing algorithms that allow machines to learn based on reinforcements. Reinforcement learning provides decision rules based on learning from partial, implicit, and delayed feedback.
This is particularly useful in sequential decision-making tasks where a machine repeatedly interacts with the environment or users. Reinforcement learning is well known for its successes in robotic control, autonomous vehicles, game playing, and chat agents. In business, economics, and the social sciences, there is a recent explosion of applications including temporal difference learning and adaptive experiments.
Application
The deadline to apply for a seat in this free course is April 24, 2025.
As the number of participants is limited, and to ensure the best fit between the course’s content and its participants, we will ask you a few questions in the application form about your background and interest in the course. We will review all applications after the deadline and notify you of the outcome by mid-May. You will get access to the course materials at the latest one week before the first meeting as you are expected to review the videos and materials of the first unit before the first meeting.
The course is open to all researchers. However, as part of our role within the consortium for business, economics, and related data, priority will be given to researchers working in these fields.
The course is completely free of charge. As part of this opportunity, we kindly ask all participants to actively contribute to our evaluation process. This helps us to continuously improve the course.
Topics
- Understanding sequential decision problems and the exploration-exploitation dilemma: multi-armed bandits
- Algorithms for earning while learning: epsilon-greedy, upper confidence bound, Thompson sampling
- Modelling the state-space: Markov processes and Markov decision processes
- Solution methods for dynamic programming: Bellman equation, value iteration and policy iteration
- Learning methods: temporal-difference learning and deep learning
- Many applications from business, economics, and social sciences
Format
This is an online course.
- Week 1:
- Multi-armed bandits and learning algorithms. Watch video lectures to learn about sequential decision problems and the exploration-exploitation dilemma. Application: How to maximize survey responses (~60 min).
- Interactive Online Session (~60 min).
- Week 2:
- Solution methods for dynamic programming. Watch video lectures to understand and apply value iteration methods. Application: How to make your firm rich and famous (~60 min).
- Interactive Online Session (~60 min).
- Week 3:
- Learning in the state-action-space. Watch video lectures about relevant theory and demonstrations of Monte Carlo and temporal difference learning. Application: Cliff Walking (~60 min).
- Interactive Online Session (~60 min).
- Week 4:
- Deep Learning. Watch video lectures about how neural networks work and how to apply them to learning problems (~60 min).
- Interactive Online Session (~60 min).
Weekly Meetings
The course includes 4 live Online Meetings, in which you will discuss the week’s contents with the instructor and fellow participants:
Meeting 1: June 09, 4:30pm-5:30pm CEST
Meeting 2: June 16, 4:30pm-5:30pm CEST
Meeting 3: June 30, 4:30pm – 5:30pm CEST
Meeting 4: July 07, 4:30pm – 5:30pm CEST
Prerequisites
Anyone with an interest in learning about reinforcement learning is welcome to take the course. It is designed for graduate students in their second year or beyond (advanced master students, advanced PhD students) to support them during their research phase.
Required: A Zoom account to participate in the online meetings.
Recommended: Knowledge of basic statistics and previous experience with any programming language is helpful, but not required.
About the Instructor
Prof. Dr. Davud Rostam-Afschar is a professor at the University of Mannheim and academic director of the German Business Panel. He has previously held visiting research positions at UC Berkeley and Harvard University, and has conducted research and taught at the University of Hohenheim, the Free University of Berlin and the University of Potsdam. He is a consultant to the European Commission and the OECD. His research focuses on macroeconomics, public economics, labor economics, and industrial economics, in particular the analysis of regulation of markets, fiscal and monetary policy. He has also a research interest in econometrics and survey methods.
