Multi-agent Learning Seminar

Time: 

 

Lectures will take place every other Monday, 14:00-16:00 at Pleinlaan 9, 3rd floor, room 3.31. 

 

 

Prerequisites: 

Either Machine learning or Techniques of AI.

You should be able to understand gradient descent and backpropagation, basic ML concepts on regression and classification, loss functions, linear and non-linear models, test error and overfitting etc. You can follow the extra resources provided in this page for a refresher.

 

Description: 

Despite the "multi-agent" title, the course will serve as an introduction to the basic concepts of single agent RL and slowly build up to advanced concepts leading to current state-of-the-art methods and Deep RL.

 

Syllabus

  • Introduction and course Overview
  • Markov Decision Processes
  • Dynamic Programming
  • Value Function Approximation
  • Policy Gradients
  • Exploration Exploitation
  • Detour: Deep Learning, Neural Nets and Convnets
  • Deep RL: Value Based
  • Deep RL: Policy Based
  • The future and open problems

 

Examination: 

Students will be evaluated based on following criteria:

  • Course participation (each student will have to hand in solutions to at least 2 course exercises + participation in discussions)
  • Learning agent project (2nd semester)
  • Project defense & report (2nd semester exam period)

 

Schedule: 

Class dates for 1st Semester:

  • 29/10/2018
  • 12/11/2018
  • 26/11/2018
  • 10/12/2018

 

Material: 

 

Basic Reinforcement Learning book:  

Reinforcement Learning: An Introduction 2nd edition

Other resources

For neural networks material

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.