A Racing Algorithm for Configuring Metaheuristics
GECCO '02 Proceedings of the Genetic and Evolutionary Computation Conference
Learning evaluation functions to improve optimization by local search
The Journal of Machine Learning Research
Stochastic Local Search: Foundations & Applications
Stochastic Local Search: Foundations & Applications
Fine-Tuning of Algorithms Using Fractional Experimental Designs and Local Search
Operations Research
Hi-index | 0.00 |
This paper formalizes the problem of choosing online the number of explorations in a local search algorithm as a last-success problem. In this family of stochastic problems the events of interest belong to two categories (success or failure) and the objective consists in predicting when the last success will take place. The application to a local search setting is immediate if we identify the success with the detection of a new local optimum. Being able to predict when the last optimum will be found allows a computational gain by reducing the amount of iterations carried out in the neighborhood of the current solution. The paper proposes a new algorithm for online calibration of the number of iterations during exploration and assesses it with a set of continuous optimisation tasks.