Multi-agent Learning Seminar

Time: 

Mondays, 14h00 CoMo Lab (10G711)

[Important notification (by lecturer, Diederik M. Roijers):
Dear students,

I sincerely apologise for the delay in the start of the course. Due to a death in my close family, I have not been able to work. I am starting again, and will keep you updated on when the course will start. We will post the start date of the course online in the coming week. Please be assured that all the lectures of the course will in fact be given, and the course will not be shortened.

In the meantime, could you please send me an e-mail at droijers [at] ai [dot] vub [dot] ac [dot] be, so that I know who has registered for the course? I will then send you a short questionaire about your interests and background. As we seem to have a small number participants this year, we can tailor the material to suit your needs and interests better when we have this information.

With kind regards and till soon,

Diederik M. Roijers (AI laboratory - VUB)

]

 
Prerequisites: 

Either Machine learning or Techniques of AI.

Description: 

This course will cover basic as well as advanced concepts of Reinforcement Learning

Introduction
Bandits
MDPs: dynamic programming
MDPs: Monte Carlo methodes
MDPs: temporal difference methods
MDPs: Model-based learning
Prioritised Sweeping
RL Learning Theory
Multi-agent learning: cooperative
Multi-agent learning: Markov Games
Multi-objective planning/RL
Deep Reinforcement Learning: DQL en double DQL
Deep Reinforcement Learning: A3C

Examination: 

Students will be evaluated based on following criteria:

  • Course participation (presentation of paper + participation in discussions + quiz)
  • Learning agent project (2nd semester)
  • Project defense & report (2nd semester exam period)

 

Material: 

 

Basic Reinforcement Learning book:  

Reinforcement Learning: An Introduction

 

Software:

  • The OpenAI Gym will be use as a testbed for learning algorithms (if you do not program your own environment, which is also allowed).
  • Experiments will be coded in Python. We recommend installation of the Anaconda framework to get all the scientific computing libraries.
  • User manual to use the project starter code on the hydra cluster.