Low budget black-box optimization is a relevant topic in many practical applications with expensive objective functions or tight real-time constraints. Recently, there has been a growing interest in addressing combinatorial permutation problems in a low budget and black-box scenario. In this context, most of the previously proposed algorithms learn a probabilistic model which guides the search by trying to somehow indicate the most effective areas of the permutation search space. However, the large size and the inherent discontinuity of the permutation space may lessen the effectiveness of this approach when a low, or very low, budget of evaluations is considered. Moving from this consideration, in this work we present a simpler elitist trajectory-based algorithm for low budget black-box optimization of permu-tation problems. The proposed algorithm, namely FAT-RLS, is based on three core ideas: a randomized local search scheme, an adaptive perturbation strength and the use of a tabu structure. A series of experiments held on commonly adopted benchmark problems clearly shows that FAT-RLS obtains better or compara-ble effectiveness with respect to the previous proposals. Moreover, its negligible computational overhead is of particular interest in mission critical situations where tight real-time constraints have to be matched.
File in questo prodotto:
Non ci sono file associati a questo prodotto.