Automatic algorithm configuration methods have shown that automatically tuning the parameters of optimization algorithms may lead to much better results, while at the same time saving significant effort to the human designer. Moreover, the potential of these methods to handle a large number of numerical, categorical and conditional parameters opens the door to more powerful applications. One of these applications is the automatic design of metaheuristics. In particular, our recent work has shown that it is possible to use an automatic configuration tool, such as irace, to generate hybrid local search algorithms. The design space of the hybrid local search algorithms is given as a grammar, from which particular algorithms may be instantiated. We convert this grammar description to a parameter space, which can be tuned for a particular problem by means of  an automatic configuration tool. The resulting system allows a human designer to automatically find the best hybrid local search algorithm for a particular problem among thousands of potential algorithm designs just by implementing a few problem-specific components.

Place: KU Leuven KULAK, E. Sabbelaan 53, 8500 Kortrijk

Time: Friday April 11, 2014, 2 pm.

By Manuel López-Ibáñez IRIDIA laboratory in the Université Libre de Bruxelles (ULB)