Look Ahead Features in Metaheuristics

In many meta-heuristics greedy choices are performed and usually measured by means of an immediate gain eventually leading to suboptimal choices. Usually, the measure of this gain as well as a mechanism to guide the choices given the evaluations is performed rather shortsighted. Therefore, especially in the artificial intelligence community researchers have thought about looking ahead (see, e.g., [19]). In that sense a heuristic measure may incorporate intelligent mechanisms to evaluate certain decisions or moves not based on an immediate gain but on the outcome after, say, a few iterations. Recently, related approaches have come up that have applied this concept as meta-heuristics for combinatorial optimization problems [4, 7]. Here we highlight the pilot method of [6, 7] (where pilot may stand as an acronym for “Preferred Iterative LOok ahead Technique” or “Perform Improved Look-ahead with Objective-value Tests”) as a look ahead strategy and show that it may be seen as a meta-heuristic to guide subordinate heuristics for determining solutions of enhanced quality. The key approach within the pilot method is that of repetition. That is, once a single heuristic is known, it is repeated over and over again without solely restricting to greedy choices.