USING A CONFLICT-BASED METHOD TO SOLVE DISTRIBUTED CONSTRAINT SATISFACTION PROBLEMS

Samaneh Semnani, Kamran Zamanifar, Ph Student
unpublished
A broad range of AI and multi-Agent problems fall in to the Distributed Constraint Satisfaction Problems category. Many of the problems in this domain are real-world problems. This fact makes DisCSPs an effective area of research. Considering all of the efforts that have recently been accomplished for solving these kinds of the problems, the most successful algorithm proposed is Asynchronous Partial Overlay (APO), which is a mediation-based algorithm. APO tries to solve the problem first by
more » ... roblem first by dividing the whole problem in to smaller portions and then solving these sub-problems by choosing some agents as mediators. This paper presents a new and effective strategy to select these mediators; moreover, it introduces two new expansion algorithms of APO that use this new strategy. These algorithms are called MaxCAPO and MaxCIAPO. The chief idea behind this strategy is that the number of mediators' conflicts (violated constraints) impacts directly on their performance. Experimental results show that choosing the agents which have the most number of conflicts as mediators not only leads to a considerable decrease in APO complexity, but also can decrease the complexity of the other extensions of the APO, such as IAPO algorithm. The results of using this conflict-based mediator selection strategy show a rapid and desirable improvement, in comparison with APO and IAPO, over various parameters such as the message and runtime complexities.
fatcat:s47qgmjrufabze2u4ib5yi27ua