TY - GEN
T1 - Algorithms for Solving Distributed Constraint Satisfaction Problems (DCSPs)
AU - Solotorevsky, Gadi
AU - Gudes, Ehud
N1 - Publisher Copyright:
Copyright © 1996 AAAI (www.aaai.org). All rights reserved.
PY - 1996/1/1
Y1 - 1996/1/1
N2 - This paper investigates Constraint Satisfaction Problems (CSPs) that are distributed by nature, i.e., there is a division of the CSF into sub components (agents) that are connected via constraints, where each subcomponent includes several of the CSF variables with the constraints between them. We call such a problem a Distributed CSF (DCSF). In this paper we give a formal deñnition of DCSFs and present four algorithms for solving them. Two of the algorithms are based on the difference between the difficulty of solving the internal constraints in the CSF components (we call them the peripheral components) of the DCSF and the difficulty of solving the constraints between the different CSPs (the central component). The two other algorithms use local and global views of the DCSF respectively. All the algorithms permit the use of different techniques (CSF, knowledge based, and operation research algorithms) in solving each of the problem components. We probe that as long as all the selected techniques are sound and complete, our algorithms are sound and complete. The algorithms were tested in a real distributed environment; the results show that when there is a difference between the difficulty of solving the peripheral components and the central one, taking advantage of it may reduce significantly the amount of work (constraint checks and message passing) needed for solving the DCSF.
AB - This paper investigates Constraint Satisfaction Problems (CSPs) that are distributed by nature, i.e., there is a division of the CSF into sub components (agents) that are connected via constraints, where each subcomponent includes several of the CSF variables with the constraints between them. We call such a problem a Distributed CSF (DCSF). In this paper we give a formal deñnition of DCSFs and present four algorithms for solving them. Two of the algorithms are based on the difference between the difficulty of solving the internal constraints in the CSF components (we call them the peripheral components) of the DCSF and the difficulty of solving the constraints between the different CSPs (the central component). The two other algorithms use local and global views of the DCSF respectively. All the algorithms permit the use of different techniques (CSF, knowledge based, and operation research algorithms) in solving each of the problem components. We probe that as long as all the selected techniques are sound and complete, our algorithms are sound and complete. The algorithms were tested in a real distributed environment; the results show that when there is a difference between the difficulty of solving the peripheral components and the central one, taking advantage of it may reduce significantly the amount of work (constraint checks and message passing) needed for solving the DCSF.
UR - http://www.scopus.com/inward/record.url?scp=10044250837&partnerID=8YFLogxK
M3 - Conference contribution
AN - SCOPUS:10044250837
T3 - Proceedings of the 3rd Artificial Intelligence Planning Systems Conference, AIPS 1996
SP - 191
EP - 198
BT - Proceedings of the 3rd Artificial Intelligence Planning Systems Conference, AIPS 1996
PB - AAAI press
T2 - 3rd International Conference on Artificial Intelligence Planning Systems, AIPS 1996
Y2 - 29 May 1996 through 31 May 1996
ER -