Multi-agent Learning (Learning Dynamics)

Professor Ann Nowé and Professor Tom Lenaerts
Teaching Assistants: Mihail Mihaylov, Peter Vrancx

Course together with the ULB. Information available here

First Semester: Thursdays at 13:00 14:00

Place: 10F720

The course addresses two general areas of research : individual-based learning and social learning in populations.

The first part focusses on learning through experience, of which reinforcement learning is the standard example. We start from a single agent setting and introduce reinforcement learning as a model free approach to dynamic programming. Then we have a look at the interplay of multiple learning agents in the same environment. For this purpose and for the following part on evolutionary dynamics, basic concepts of Game Theory are introduced.

The second part provides an introduction to the principles of learning by imitation, modelled through evolutionary dynamics. It will explain what evolution is and how games can be used to model interactions between individuals in a population. It will show how these models can be used to study the evolution of cooperation in social dilemmas, the evolution of conventions like language or even the dynamics of cancer.

The course concludes with a project which can include, for those who are interested, experiments using the Khepera robots.

Additional reading will be available later.

The most recent course notes will become available before each lecture.

• Game Theory part 1
• Game Theory part 2
• Introduction
• Part 1
• Part 2
• Part 3
• Part 4
• Part 5
• MAS RL
• Cultural Learning and Evolution
• Evolution of Cooperation
• Sparse Interactions in Multi-Agent Reinforcement Learning

You can also download a pdf version of this schedule.

TBA

• During the course sessions, students will receive small assignments, of which the results are returned as a report during one of the following sessions.
• Students will have to select a project that needs to be finished before the exam period in january.
• The quality of the project and the quality of the defense will determine the final points for this course.

Mail your solutions to mmihaylo [AT] vub.ac.be, or hand-in a paper copy to Prof. Nowé. For electronic versions please provide a single (self-contained) *.PDF or *.DOC file. Put the assignment number and your name in the filename.

Assignments are personal. No groupwork allowed!

• Assignment 1 (deadline: 13.10.2011)
• Assignment 2 (deadline: 20.10.2011 24.10.2011)
• Assignment 3 (deadline: 17.11.2011)
• Assignment 4 (deadline: 01.12.2011)

Sample solutions:

• Assignment 1
• Assignment 2
• Assignment 3
• Assignment 4

Project proposals are now available.

The exam will take place on 16 January 2012 in room 10F720.
Groups 2,3,5 and 7 should be present from 9:30 until 12:00.
Groups 1,6 and 9 should be present between 13:30 and 15:30.
All groups are allowed to participate in both the morning and afternoon sessions.