Workshop Cooperative Multi-Agent Learning

 

 

A Workshop at the ECML-PKDD 2005 Conference

 

3 October 2005

 

 

Workshop Programme

 

9:00-9:15

Welcome and opening

9:15-10:00

Invited talk by Gusz Eiben (Free University Amsterdam)

10:00-10.30

Break

10:30-11:00

Link: The impact of social networks on multi-agent recommending systems

11:00-11:30

Ciesielski: Efficient Cooperation via Conservative Reconfiguration of  Agents Coalitions

11:30-12:00

Riachi, Asadpour, Siegwart : Cooperative Learning for very long learning tasks: a society-inspired approach to persistence of knowledge

12:00-12:30

Daneshvar: Is environment a common memory?

12:30-14:00

Lunch

14:00-14.30

Campos, Brazdil: A Multi-Agent Approach to Technological Imitation and Collective memory

14.30-15:00

 
Nunes, Oliveira: Communicating During Learning

15:00-15:30

Break

15:30-16:00

Bakker, Steingrover, Schouten, Nijhuis, Kester: Cooperative multi-agent learning of traffic lights

16:00-16:45

Discussion: the future of cooperative multi-agent learning


           

 

Cooperative Multi-agent Learning is part of Machine Learning that emphasizes the joint behaviors of learning agents in environments with some degree of autonomy. In most such environments there are constraints placed on the degree to which any agent may know what other agents know, or on their communication capabilities, such that the system must have distributed control and cannot be solved with a master-slave model via a single master agent.

The presence of large numbers of agents, increasingly complex agent behaviors, partially observable environments, and the mutual adaptation of agent behaviors make the learning process a challenging one. These problems are further complicated by noisy sensor data, local bandwidth-limited communication, unplanned faults in hardware agents, and stochastic environments.

A closely related area is Distributed Data Mining in which the main task of agents is to construct a model of the environment. Limits on capacity, costs or privacy limitations require a distributed approach to Data Ming. A wide range of Machine Learning issues appear in a new form in multi-agent settings, such as constructive induction, relevance learning, ensemble learning, learning bias, overfitting.

The problems of coordination, communication and integration of information can be approached from different perspectives: human learning, game theory and economics, distributed systems, distributed knowledge representation and the semantic web, development and evolution. Computationally oriented studies of Cooperative Multi-Agent Learning are also welcome.

Practical setting in which multi-agent learning is adapted are: robot soccer, mining large datasets in astronomy, distributed data mining in transportation, finance, user modelling, the semantic web.

The goal of this workshop is to bring together researchers from diverse areas including multi-agent systems, cognitive science, distributed computing for data mining, collaborative data mining to review work in this emerging area and to articulate a research agenda for the coming years.

Topics of interest include, but are not restricted to:

  • Co-adaptation, co-evolution and modelling
  • Collaborative Reinforcement Learning
  • Swarm and social learning methods
  • Evolutionary game theory
  • Multirobot learning
  • Learning to communicate
  • Protocols for distributed learning
  • Multi-agent and Distributed approaches to Data Mining
  • Distributed Data Mining over the Grid
  • Evolution of common language, ontology and culture

 

Organizers:

 

Maarten van Someren and Nikos Vlassis (University of Amsterdam, The Netherlands)

 

Committtee:

 

Maarten van Someren (University of Amsterdam, Netherlands)

Pavel Brazdil (University of Porto, Portugal)

Pete Edwards (Aberdeen, Scotland)

Nikos Vlassis (University of Amsterdam, Netherlands)

Marco Wiering (Universiteit Utrecht, Netherlands)

Christian Lebiere (Micro Analysis and Design, Inc., USA)

Eugenio Oliveira (University of Porto)

Ron Sun (Rensselaer Polytechnic Institute, USA)

Vasant Honavar (IOWA State University)

Jürgen Franke (DaimlerChrysler AG, Germany)

Wim Wiegerink (Radboud University Nijmegen, Netherlands)

Edwin de Jong (Universiteit Utrecht, Netherlands)

 

 

Submissions:

The papers must be in English and should be formatted according to the Springer-Verlag Lecture Notes in Artificial Intelligence guidelines. Authors instructions and style files can be downloaded at http://www.springer.de/comp/lncs/authors.html. The maximum length of papers is 12 pages. The workshop proceedings of ECML and PKDD will be published as workshop notes at the conference. Authors keep the copyright. The possibility of publishing a revised version of the papers as special issue of a journal is currently being explored. Simultaneous submission to other workshops and conferences is allowed, provided this is clearly indicated on the submission.

Papers can be submitted electronically by sending them to: maarten@science.uva.nl

 

 

Important dates:


Workshop paper submission deadline: 25 July 2005 PASSED
Workshop paper acceptance notification: 15 August 2005 PASSED
Workshop paper camera-ready deadline: 5 September 2005 PASSED

Workshop: Monday 3 October 2005

 

 

Contact information:

 

Maarten van Someren

Informatics Institute

University of Amsterdam

Kruislaan 419

1098 VA Amsterdam

The Netherlands

email: maarten@science.uva.nl

tel: +31 20 525 6791