Adaptive Compromises in Distributed Problem Solving

ABSTRACT The ERA method is a multi-agent oriented method for solving onstraint satisfa tion problems. During problem solving, distributed agents in ERAmake de isions based on their own lo al environments. Although ea h agent has multiple lo al variables and the rea tive behaviors of the agent are lo al, the system an rea h a solution state. In this paper, we introdu e agent ompromises in solving a GCP (Graph Coloring Problem) by using ERA method. The edges in a GCP are transformed into two lasses of onstraints: lo al onstraints and neighbor onstraints. In order to solve a given problem, we need to minimize the number of unsatis ed neighbor onstraints. When the system gets stu k in a lo al minimum in some time steps, a ompromise of two neighboring agents that have ommon violated onstraints is made. The ompromise may eliminate the original neighbor violations and a t as a single agent to determine its own de ision. The ompromise will be abandoned after the rea tion is nished. We have ondu ted some experiments to evaluate various ompromise strategies. Experimental results show that ompromise strategies help to improve the performan e of GCP solving.