course_description.txt

Modulkatalog Lernziele: 
    Die Studierenden sind in der Lage Problemstellungen als Reinforcement Learning Aufgabe zu formulieren und Lösungsansätze aus aktuellen Forschungsergebnissen zu entwickeln. 
    Dazu haben sie einen Überblick über die Forschungslandschaft mit Schwerpunkt Meta Reinforcement Learning gewonnen. 
    In einen abschließenden Projekt haben die Studierenden gelernt diese Konzepte selbständig weiterzuentwickeln und praktisch umzusetzen.

Modulkatalog Stoffplan: Erweiterungen des MDP Frameworks, Aktuelle RL Algorithmen, RL in der Praxis, Meta RL, RL Lernen

Modulkatalog Vorkenntnisse: Reinforcement Learning

Modulkatalog Literaturempfehlungen: -


Description Website:
    Reinforcement Learning (RL) in principle offers a highly interesting path for AI problem solving: entwining data collection and model learning to find novel solutions.
    While this has been successfully done in games, robotics and even some natural sciences like physics or chemistry, RL solution strategies are to date brittle and often lack generalization capabilities. 
    Therefore RL research has been trying to find better algorithms -- through improving the algorithms themselves, configuring them more efficiently or learning to support them in their tasks.
    This course will cover these ideas and you see how RL and MetaRL can be combined to solve RL problems. 
    The accompanying seminar session will offer opportunities for you to discuss how the ideas from the lecture can be applied in non-standard RL scenarios like continual learning. 
    At the end of the semester, you will apply these skills to drafting a research proposal improving one such setting for the exam.

Recommended pre-requisites:
    The contents of our lecture "Reinforcement Learning" are mandatory for this course. We will *not* have time to recap these basics, you'll be expected to know the concepts and algorithms from this lecture and be able to implement them!

Lecture topics include:
    - Extensions to Markov-Decision Processes
    - The state of RL algorithms
    - RL in research & practice
    - MetaRL
    - Learning to Reinforcement learn