Department of Information and Computing Sciences

Departement Informatica Onderwijs
Bachelor Informatica Informatiekunde Kunstmatige intelligentie Master Computing Science Game&Media Technology Artifical Intelligence Business Informatics

Onderwijs Informatica en Informatiekunde

Vak-informatie Informatica en Informatiekunde

Multi-agent learning

Website:website containing additional information
Course code:INFOMAA
Credits:7.5 ECTS
Period:period 4 (week 17 through 26, i.e., 24-4-2017 through 30-6-2017; retake week 28)
Timeslot:C
Participants:up till now 53 subscriptions
Schedule:Official schedule representation can be found in Osiris
Teachers:Dit is een oud rooster!
formgrouptimeweekroomteacher
lecture   Tue 13.15-15.0017-23 BBG-023 Gerard Vreeswijk
 
24 RUPPERT-116
25 BBG-023
26 RUPPERT-116
Thu 9.00-10.4518-19 BESTUURS-LIEREGG
20 RUPPERT-ROOD
22-23 RUPPERT-ROOD
24 BESTUURS-LIEREGG
25-26 RUPPERT-ROOD
Contents:This seminar discusses forms of machine learning that typically occur in multi-agent systems. Topics: learning and teaching, fictitious play, rational learning, no-regret learning, multi-agent reinforcement learning, evolutionary learning.

The seminar assumes knowledge of probability theory, and superficial knowledge of game theory and machine learning.
Literature:Available through the course site. Download and print of material is a responsibility of the student.
Course form:Seminar.
Exam form:
  • Participation 10%
  • Assignments 15%
  • Summaries 30%
  • Presentation 15%
  • Paper 30%
  • Minimum effort to qualify for 2nd chance exam:Om aan de aanvullende toets te mogen meedoen moet de oorspronkelijke uitslag minstens 4 zijn.
    Description:Multi-agent learning (MAL) studies software agents that learn and adapt to the behaviour of other software agents, that themselves adapt to the behaviour of other software agents. The presence of other learning agents complicates learning, which makes the environment non-stationary (a situation of learning a moving target) and non-Markovian (a situation where not only experiences from the immediate past but also earlier experiences are relevant). With adaptive agents it also becomes less beneficent to only adapt to the behaviour of other agents, on the pain of being exploited by more steadfast agents that do not follow but instead impose their strategy on others. Important topics of adaptive agents include statistical learning and single-agent reinforcement learning. Important topics of MAL include (evolutionary) game theory, fictitious play, gradient dynamics, no-regret learning, multi-agent reinforcement learning (MinMax-Q, Nash-Q), leader (teacher) vs. follower (learner) adaptation, and the emergence of social conventions. Examples of domains that need robust MAL algorithms include manufacturing systems (managers of a factory coordinate to maximise their profit), distributed sensor networks (multiple sensors collaborate to perform a large-scale sensing task under strict power constraints), robo-soccer, disaster rescue (robots must safely find victims as fast as possible after an earthquake) and recreational games of imperfect information such as poker. Indeed, poker and simplified forms of poker are an important topics of research in multi-agent learning.
    wijzigen?