Abstract

Finding a compromise between the exploration of new opportunities that could yield excellent performances and the exploitation of existing solutions through local refinements represents a major challenge in different sectors. In fact, although the search for improved solutions may be costly and time consuming in the short term, its effects could have a big impact in the long term. Conversely, exploitation is likely to be beneficial in the short term, but might have a catastrophic effect in the long term. With respect to evolutionary computation, several approaches attempt to address the issue from different perspectives. In this work, we analyze the exploration-exploitation dilemma in problems where neutrality --- the condition according to which the search space consists of vast areas accessible through mutations that do not jeopardize the survival chances --- is a distinctive feature. Specifically, we present the results achieved in two benchmark problems: (i) function optimization and (ii) 5-bit parity. Moreover, a novel method mixing exploration and exploitation, called SSSHC*, is introduced and compared with two other algorithms. The experiments reported in this paper indicate that the SSSHC* finds remarkably better solutions than other methods in the optimization of function tests and is competitive with respect to the parity problem. Overall, the outcomes show how exploration and exploitation can be effectively combined, but the strategy for doing so is task-dependent.

Statistics