Reinforcement Learning

• Organized by: Prof. Joschka Boedecker, Gabriel Kalweit and Maria Huegle
• Lecture: HISinOne number: 11LE13V-1141
• Language: English
• Lecture:
• Friday 14:00 - 16:00 (s.t.)
• Building 101, Seminar 01-009/013
• First session: 2019-10-25 14:00 (c.t.)
  
• There will be no exercise sessions, but solutions for the exercises will be posted on ILIAS.

Overview:

The lecture deals with methods of Reinforcement Learning that constitute an important class of machine learning algorithms. Starting with the formalization of problems as Markov decision processes, a variety of Reinforcement Learning methods are introduced and discussed in-depth. The connection to practice-oriented problems is established throughout the lecture based on many examples.

All slides and exercises can be found in ILIAS.

Exam:

There will be oral exams, the dates are March 23-25. The first six minutes will be about the project, talk and discussion. The project is designed as an exercise sheet for the last three weeks of the lecture. The final grade will be a weighted average with a weight of 1/3 for the project and 2/3 for the questions about the rest of the lecture and exercises. You can work on the project in groups of up to 3 people.

You can choose to implement and apply any reinforcement learning algorithm (from the lecture or beyond) to solve the given problem. The evaluation should at least include learning curves (i.e. the return over time) of your chosen approach and settings. You can additionally think of your own metric and evaluate that as well. It is important that your evaluation builds the basis for discussion and scientifically analyzes which are the important aspects and characteristics of your approach – your talk has to highlight your findings in a convincing manner. You can prepare two slides, one with your approach and one with results (prepare as many backup slides for the discussion as you want). You have to submit your slides and code on February 14 latest.