AAAI-2002 Spring Symposium
on
COLLABORATIVE LEARNING AGENTS
March 25-27, 2002
Stanford University, Stanford, CA


Organizing committee members:


DESCRIPTION

Recent advances in the Multi-Agent Systems (MAS) field have generated optimism that widely applicable solutions to large, distributed problems may be at hand. However, before the field can deliver on that promise, the challenge of how to control such systems to address a pre-specified goal (e.g., minimize throughput of packets in data routing, win the game in soccer) in a decentralized, adaptive manner with minimal detailed hand-tuning needs to be met.

In this workshop we focus on two crucial properties that would allow a MAS to meet those challenges:

The first property is crucial in large problems (e.g., internet routing), and inherently distributed problems (e.g., planetary exploration rovers, constellations of satellites), in that it enables a modular approach to the problem. Without it, either each agent will attempt to solve the full problem (unlikely to succeed in any but the simplest problems) or agents will get in one another's way as they frustrate one another, preventing the system from reaching desirable states.

The importance of the second property lies in how the agents interact with one another and the environment. Because both the environment and the response of other agents to changes in that environment will modify the "background" state one agent perceives before choosing its actions, it is imperative that adaptivity be built in to those agents. Without this flexibility, only detailed hand-tailoring (an option not available in most large, interesting problems) will provide satisfactory behavior. Furthermore, the interaction structure among the agents needs to be adaptive, to allow the agents to fully exploit opportunities in a changing environment (e.g., form/dissolve teams).

There are many fields that have addressed aspects of these issues, including economics, and game theory. However, there are major differences in both the approach such fields take, and the set of assumptions that form the basis of those fields. For example, the components of a multi-agent system have many degrees of freedom that human beings do not. Due to this freedom, a MAS designer has a much larger "set of tools" than economists have. Furthermore, while game theory has established a strong theoretical basis for analyzing the equilibrium behavior of systems and how various equilibrium states relate to one another, there is little work on off-equilibrium behavior.

Therefore, neither the direct application, nor the simple extension of principles borrowed from those fields are likely to provide the theoretical underpinnings of teamwork and adaptivity in multi-agent systems. Our focus in this workshop will be to address the design of systems that are intended to solve large distributed computational problems with little to no handtailoring through the collective and adaptive behavior of the agents comprising that system.


Topics of interest

To address these issues, and stimulate discussion on related issues, we encourage submissions in the following areas:

Submission Requirements

E-mail the URL of either a
Papers and statements of interest must be in one of the following formats: postscript, pdf, HTML.

Direct all questions and inquiries to the co-chairs:
Peter Stone
AT&T Labs -- Research
Rm. A273
180 Park Ave.
Florham Park, NJ 07932
pstone@cs.cmu.edu
http://www.cs.cmu.edu/~pstone
phone: (973) 360-8333
fax : (973) 360-8970

Kagan Tumer
NASA Ames Research Center
Mail Stop 269-4
Moffett Field, CA 94035-1000
kagan@ptolemy.arc.nasa.gov
http://ic.arc.nasa.gov/~kagan/
phone: (650) 604-4940
fax : (650) 604-3594


Important Dates

Submissions due: October 15, 2001 (extended from October 5th)
Acceptance/rejection notifications sent: November 30th, 2001
Camera-ready copies due: January 7, 2001
Symposium: March 25-27, 2002