6 page gzipped Postscript file (51K)

Next: Introduction

Multiagent Learning through Collective Memory

Andrew Garland and Richard Alterman
Computer Science Department
Brandeis University
Waltham, MA 02254
{aeg, alterman}@cs.brandeis.edu

Abstract:

Collective memory is a mechanism through which a community of agents can learn improved patterns of cooperative activity to take advantage of regularities in their domain of problem-solving activity. Collective memory is defined as the breadth of procedural knowledge the community acquires through experience when interacting with each other and the world. This knowledge may address many aspects of multiagent problem solving, but we will focus on two aspects: learning cooperative procedures and learning agents' capabilities. We demonstrate how collective memory reduces the amount of total effort and communication needed by a community of agents to solve problems. Finally, we discuss how this model of behavior is better suited for our domain of interest than related work.





Andrew Garland
Thu Apr 9 11:37:41 EDT 1998