Heuristic Revision by Heuristic Space Exploration

Heuristics are often used to solve complex problems. Indeed, such problem-specific knowledge, when pertinent, helps to efficiency find good solutions to complex problems. Unfortunately, acquiring and maintaining a heuristic set can be fastidious. In order to face this problem, an approach consists in revising the heuristic sets by means of experiments. In this paper, we are interested in a specific revision method of this type based on the exploration of the heuristic space. The principle of this method is to revise the heuristic set by searching among all possible heuristics the ones that maximize an evaluation function. In this context, we propose a revision approach, dedicated to heuristics represented by production rules, based on the reduction of the search space and on a filtered local search. We present an experiment we carried out in an application domain where heuristics are widely used: cartographic generalization.