Toward an Efficient Exploration of Fitness Landscapes
Résumé
Within local search algorithms, descent methods are rarely studied experimentally. However,
these search techniques are the basis of many modern metaheuristics and have an influence on the
ability of an algorithm to achieve good solutions of a fitness landscape. Through a large empirical study
of classic runs, we show that certain ideas about descents methods are false. These results indicate
that it is possible to make a descent ’intelligent’ and lead to better solutions, regardless of the problem
addressed.