Selected papers at Learning and Intelligent Optimization: Third International Conference, LION 3, Trento, Italy, January 14-18, 2009

Beam-ACO algorithms are hybrid methods that combine the metaheuristic ant colony optimization with beam search. They heavily rely on accurate and computationally inexpensive bounding information for choosing between different partial solutions during the solution construction process. In this work we present the use of stochastic sampling as a useful alternative to bounding information in cases were computing accurate bounding information is too expensive. As a case study we choose the well-known travelling salesman problem with time windows. Our results clearly demonstrate that Beam-ACO, even when bounding information is replaced by stochastic sampling, may have important advantages over standard ACO algorithms.

Peer Reviewed