The basic idea of VNS is the change of neighbourhoods in the search for a better solution. VNS proceeds by a descent method to a local minimum exploring then, systematically or at random, increasingly distant neighbourhoods of this solution. Each time, one or several points within the current neighbourhood are used as initial solutions for a local descent. The method jumps from the current solution to a new one if and only if a better solution has been found. Therefore, VNS is not a trajectory following method (as Simulated Annealing or Tabu Search) and does not specify forbidden moves. In this work, we show how the variable neighbourhood search metaheuristic can be applied to train an artificial neural network. We define a set of nested neighbourhoods and follow the basic VNS scheme to carry out our experiments

Abstract. Most Data Mining tasks are performed by the application of Machine Learning techniques. Metaheuristic approaches are becoming very useful for designing efficient tools in Machine Learning. Metaheuristics are general strategies to design efficient heuristic procedures. Scatter Search is a recent metaheuristic that has been successfully applied to solve standard problems in three central paradigms of Machine Learning: Clustering, Classification and Feature Selection. We describe the main components of the Scatter Search metaheuristic and the characteristics of the specific designs to be applied to solve standard problems in these tasks.

No abstract