The Gestalt heuristic: learning the right level of abstraction to better search the optima

Nowadays, engineering process often requires the optimization of a cost: improving energetic yield, minimizing the used space, reducing the development effort, . . . Not surprisingly, research in optimization is one of the most active field of computer science. Metaheuristics are among the state-of-the-art techniques for combinatorial optimization problem. In this context, the paper addresses the following question: “Considering computation time and implementation ease constraints, is it then possible to build a system increasing the efficiency of the used metaheuristic by automatically altering the problem representation during the optimization process?”. Inspiring by the Gestalt psychology, a descriptive theory of complex cognitive processes such as vision, the paper suggests the Gestalt heuristic as a valid solution. The heuristic mechanism mainly consists in building a meta-model, an abstraction of the problem. This meta-model is used as a filter or a lens inserted between the metaheuristic and the problem representation. From an engineering point of view, the Gestalt heuristic is a promising additional mechanism for rapid prototyping and anytime optimization context. After introducing a formal and unified framework for Gestalt heuristic, the paper gives an illustrated guideline for the implementation of this newheuristic. The suggested implementation is then experimentally tested and discussed.