CSAA: A Distributed Ant Algorithm Framework for Constraint Satisfaction

May 2004

Proceedings of the Seventeenth International Florida Artificial Intelligence Research Society Conference (FLAIRS 2004)

Volume

Proceedings of the Seventeenth International Florida Artificial Intelligence Research Society Conference (FLAIRS 2004)

All Papers

In this paper the distributed Constraint Satisfaction Ant Algorithm (CSAA) framework is presented. It uses an ant-based system for the distributed solving of constraint satisfaction problems (CSPs) and partial constraint satisfaction problems (PCSPs). Altough ant algorithms have already proven to be a viable alternative to classical approaches for a variety of optimization and combinatorial problems, current ant systems work in a centralized manner. Problems where the flexibility of ant systems can be useful, usually tend to get large. Therefore a distributed solving approach is needed. We show that when the distribution is done in an appropriate manner, ant algorithms conserve their flexibility. The distribution however is not trivial. A number of difficulties (especially with relation to speed and accuracy) emerge when the centralized framework would just be distributed over multiple hosts. In this paper we address those difficulties and provide solutions. We show that with the right design decisions, a distributed ant algorithm is a viable alternative for classical approaches. When flexibility in the solving method becomes an issue (for example in dynamic problems), ant algorithms, who use an flexible decision mechanism without hard commitments, even have an advantage over traditional algorithms.