Modulbeschreibung

Deep Reinforcement Learning

ECTS-Punkte:
2
Lernziele:

n this module, we introduce the foundations of deep reinforcement learning and will apply different algorithms to solve concrete problems in robotics and/or games. 

After successful completion of this module the students:

  • know how to apply deep reinforcement learning (RL) to solve problems in robotics and automation.
  • know how to implement RL in Python using state-of-the-art libraries (for example TensorFlow, Keras)
  • know how to formally describe a task as an RL problem
  • understand what a computational-graph is and how it is used by neural networks
  • understand the link between RL and Deep Learning
  • know the limitations of RL
  • understand the role of simulators and know how to use a robotics simulator (for example openAI-Gym)
  • know how to qualitatively and quantitatively describe the progress of the training process.

Kurse in diesem Modul

Deep Reinforcment Learning:

Learn how to describe a problem:

  • describe a problem in terms of a Markov decision making task
  • Fundamentals: state, action, reward, policy, Q-value
  • Bellman equations

Learn algorithms to solve RL problems:

  • SARSA, Deep-Q (and some of their variants)
  • Backprop (Deep RL)
  • optional: model-based RL

Learn to use Python and its libraries:

  • Libraries: (for example Keras, TensorFlow)
  • Simulation: (for example: openAI-Gym)

Apply your skills in a (graded) project

  • Implementation of a project in the domain of robotics (e.g. https://gym.openai.com/envs/#robotics) or games (e.g. Connect Four)
Klassenunterricht mit 2 Lektionen pro Woche
Disclaimer

Diese Beschreibung ist rechtlich nicht verbindlich! Weitere Informationen finden Sie in der detaillierten Modulbeschreibung.