Modulbeschreibung

Reinforcement learning is one of the most exciting areas of machine learning and unfortunately, also one of the most complex. It is a machine learning (ML) paradigm that is capable of optimizing sequential decisions. RL is interesting because it mimics how we, as humans, learn. We are instinctively capable of learning strategies that help us master complex tasks like riding a bike or taking a mathematics exam. RL attempts to copy this process by interacting with the environment to learn strategies. RL is a highly mathematical topic.
This course is aimed at MSE students of data science and other technical disciplines. It is intended to be an introduction to deep RL and no prior knowledge of the subject is required. However, we do assume that readers have a basic familiarity with machine learning and deep learning as well as an intermediate level of Python programming. Some experience with tensorflow, keras is also useful but not necessary.

The students…

• learn the theory and application of models and algorithms used in Reinforcement Learning (RL)
• understand the key concepts of Deep Reinforcement Learning: Markov Process, Bellman equation, Q-Learning, TD-Learning, Deep Q-Learning, policy gradients and actor-critic methods.
• get to know a wide range of possible application examples and useful RL environments
• learn how to approach their own DRL projects and how to implement them in a working Python code.
• get the skills to research, study and understand primary sources in form of academic papers.
• get the skills in Python and Tensorflow to work on and implement practical, industrial DRL control projects.

1. Introduction to RL
2. Markov Decision Processes, Dynamic Programming, and Monte Carlo
3. Temporal-Difference Learning, Q-Learning, and n-Step Algorithms
4. Deep Q-Networks: DQN and DDQN
5. Policy Gradient Methods
6. Actor-Critic-Algorithms: synchronous and asynchronous parallelization techniques that are applicable to any of the algorithms
7. Practical Reinforcement Learning