Multi-Agent Learning Seminar / Reinforcement learning

Professor
Ann Nowé
Course description
Course Content:
  • Introduction, without any prior requirement except being able to code, to the wide world of Reinforcement Learning. This course teaches how to make artificial agents that learn by “trial and error”, suited for various kinds of simple to complicated tasks. It also covers the basics of Neural Networks, Deep Learning, Control Engineering, Stochastic Optimization and Planning.
  1. Introduction and course Overview
  2. Markov Decision Processes
  3. Dynamic Programming
  4. Value Function Approximation
  5. Policy Gradients
  6. Exploration Exploitation
  7. Detour: Deep Learning, Neural Nets and Convnets
  8. Deep RL: Value Based
  9. Deep RL: Policy Based
  10. The future and open problems
Materials:

Basic Reinforcement Learning book:  Reinforcement Learning: An Introduction 2nd edition

  • Other resources
  1. Practical Reinforcement Learning Coursera
  2. John Schulman’s and Pieter Abeel’s class: Deep Reinforcement Learning, Fall 2015
  3. Deep Reinforcement Learning and Control, CMU Spring 2017
  4. David Silver’s class: Reinforcement learning
  • For neural networks material
  1. Andrej Karpathy’s course
  2. Geoffrey Hinton on Coursera
  3. Goodfellow’s Deep Learning book
  • Software
  1. The OpenAI Gym will be use as a testbed for learning algorithms (if you do not program your own environment, which is also allowed).
  2. Experiments will be coded in Python. We recommend the use of “pip”, available on any distribution, and with every Python library available with a single “pip3 install library“. For Windows users, pip is also available with recent versions of Python 3, but in case of any problem, here is a link to the Anaconda framework.
  3. User manual to use the project starter code on the hydra cluster.
The corresponding competences:
  • Knowledge and Insight:
    The student has knowledge and insight in the domain of learning systems which allows him to possibly provide an original contribution to the domain.
  • The use of Knowledge and Insight:
    The student can combine the ideas covered in the course to obtain a suitable approach for a new problem.
  • Judgements Ability:
    The student can judge autonomously the scientific papers in this domain. 
  • Communication:
    The student can present the content of their final project to the other students and communicate his ideas on the solutions.
  • Skills:
  • The student can autonomously search, read and implement papers in this area of research. 

All detailed and official information about the course here >