|Venue and Contact:
CS Dept., University of Wales, Aberystwyth
About the Society for the Study of Artificial Intelligence
and the Simulation of Behaviour (AISB)
AAMAS contact: email@example.com
Kazakov, CS Dept., University of York, Heslington, York YO10 5DD, UK
For contacts until 12 Dec 2002:
Department of Intelligent Systems, Jozef Stefan Institute
Jamova 39, 1000 Ljubljana, Slovenia
tel. +386 1 477 3693, fax: +386 1 4251 038.
Frances Brazier, Department of AI, Free University, Amsterdam.
Niek Wijngaards, Department of AI, Free University, Amsterdam.
Ann Nowé, Free University of Brussels.
Kurt Driessens, Computer Science Department, Catholic University
Tom Holvoet, Computer Science Department, Catholic University
Saso Dzeroski, Jozef Stefan Institute, Ljubljana.
Philippe De Wilde, Imperial College, London.
Michael Schroeder, Department of Computing, City University,
Kostas Stathis, Department of Computing, City University,
Eugenio Oliveira, Department of Computing and Electrical
Engineering, University of Porto.
Enric Plaza, IIIA-CSIC, Spain.
|Call for Papers
Adaptive Agents and Multi-Agent Systems (AAMAS) is an
emerging multi-disciplinary area encompassing Computer Science, Software
Engineering, Biology, as well as Cognitive and Social Sciences.
When designing agent systems, it is impossible to foresee
all the potential situations an agent may encounter and specify the agents'
behaviour optimally in advance. Agents therefore have to learn from and
adapt to their environment. This task is even more complex when the agent
is situated in an environment that contains other agents with potentially
different capabilities and goals. Multi-Agent Learning, i.e., the ability
of the agents to learn how to co-operate and compete, becomes crucial in
The goal of this symposium is to increase awareness and
interest in adaptive agent research, encourage collaboration between ML
experts and agent system experts, and give a representative overview of
current research in the area of adaptive agents. The symposium will serve
as an inclusive forum for the discussion on ongoing or completed work in
both theoretical and practical issues.
The symposium is a continuation of AAMAS,
held as part of AISB-01 in York, March 2001, and AAMAS-2,
held in London as part of the AISB-02 in April 2002. The AAMAS
are the first scientific meetings on learning agents in the UK, and the
success in the previous two years has clearly confirmed the need for such
The symposium topic is situated at the intersection of
two areas, namely,
Adaptation/Learning and Agents, which would naturally
relate to the general convention theme: Cognition in Machines and
Animals. The symposium will focus on (but is not necessarily limited
to) the following topics:
Adaptive Mobile Agents: adaptation of and to platforms
(adaptive body rather than mind).
From Single Agent to Multi-Agent Learning:
The ability to learn is especially important for an agent when there are
other agents acting in the environment. An important open question is whether
and how single-agent learning techniques can be modified and applied in
a multi-agent setting.
Learning of Co-ordination: It
is obvious that agents in a multi-agent system need to co-ordinate their
action, whether in a co-operative or competitive manner. It is often not
feasible to develop a good co-ordination protocol from scratch, and therefore
agents need to acquire co-ordination skills by learning. Co-ordination
(and learning of it) can be studied under several different assumptions,
e.g., with/without communication or with/without mutual observation, depending
on the application area and the respective restrictions.
Learning and Communication: When several learning
agents work in a team it may be beneficial for them to cooperate not just
on the task achievement but also on the learning process itself. Clearly,
communication is an important tool for such co-operation.
Distributed Learning: The major question in this area
is how agents can learn in a collaborative way as a group. This is in contrast
to the alternative view on multi-agent learning where agents in a group
learn individually and separate theories are obtained.
Evolutionary Agents: Natural selection can be
employed to evolve in a generation of agents with inherited properties
the phenotype that best fits the agents' goals in a given environment.
The approach has been successfully applied to
social simulation and other multi-agent domains.
Emergent Organisation/Behaviour and
Studies of Complexity in Multi-Agent Systems
with Learning and Adaptation: Understanding
of how properties such as functional organisation,
adaptability and robustness can emerge from complex learning
systems. Also, studies of the computational complexity
of learning algorithms and its impact on the way in which
learning and recall are balanced in MAS.
Evolution of Individual Learning in Multi-Agent
Systems: Individual agents can use personal experience
in order to improve their performance throughout their lifespan. Alternatively,
reproduction and natural selection can be employed in a MAS to evolve agents
that are best suited for the task. The two approaches can be combined,
and natural selection used to evolve agents with an optimal learning bias.
This topic also links natural selection, language and learning through
the evolutionary search for the best language (or language bias) used for
Game-Theoretical and Analytical Approaches to Adaptive
Multi-Agent Systems: No research on learning agents will be complete
if it does not draw on the body of work in Game Theory and Systems Theory
in order to compare its results with these two areas and provide a unifying
view of the different aspects of interaction within a system, resp. among
agents, that each approach provides.
Logic-Based Learning: The ability to incorporate background
knowledge into the agents' decision-making and learning processes is arguably
essential for effective performance in complex, dynamic domains. Logic-based
learning mechanisms such as explanation-based learning and inductive logic
programming are commonly used in such situations. The MAS setup brings
in several specific issues, such as reasoning about time and in time (for
an action to be taken), the ability to communicate observations and theories,
and to cope with inaccurate or misleading information.
Learning in Reactive Agents: Learning in a setup where
the only model of the world an agent has is the world itself. (Also: learning
of control, sub-symbolic and lazy learning).
Learning for Real-Time Applications:
An agent typically performs a number of tasks, learning being just one
of them. Time complexity of learning and validity of results, trade-off
between learning (improving performance) and recall (using acquired knowledge
to perform other, non-learning, tasks) are all relevant to this topic.
Flexible Real-Time Systems (RTS) are a hot research topic, yet very little
has been done to combine results from ML and RTS areas.
Industrial and Large Scale Applications of Learning Agents:
Agent technology is already having a strong impact on various applications,
including e-commerce, entertainment, human-computer interfaces, and plant
control. Many of these applications are being equipped with machine learning
Electronic submissions of extended abstracts of
up to four pages should be submitted as postscript or PDF files to
firstname.lastname@example.org with the subject line "paper
submission" by 6 Jan 2003. Up to five keywords of
the author's choice should be listed in the E-mail rather than
the paper itself.
New Submission Deadline for Extended Abstracts:
20 January 2003
Notification re: Extended Abstracts: February 2003
Submission of Full Papers:
7 March 2003
Convention: 7 - 11 April 2003 (the exact
days and duration of the symposium are yet to be set)
Proc. of AAMAS
Proc. of AAMAS-2
Book on AAMAS (forthcoming)