Algorithm selection for black-box continuous optimization problems: A survey on methods and challenges

Abstract

a b s t r a c tSelecting the most appropriate algorithm to use when attempting to solve a black-box continuous optimization problem is a challenging task. Such problems typically lack algebraic expressions, it is not possible to calculate derivative information, and the problem may exhibit uncertainty or noise. In many cases, the input and output variables are analyzed without considering the internal details of the problem. Algorithm selection requires expert knowledge of search algorithm efficacy and skills in algorithm engineering and statistics. Even with the necessary knowledge and skills, success is not guaranteed.In this paper, we present a survey of methods for algorithm selection in the black-box continuous optimization domain. We start the review by presenting Rice's (1976) selection framework. We describe each of the four component spaces -problem, algorithm, performance and characteristic -in terms of requirements for black-box continuous optimization problems. This is followed by an examination of exploratory landscape analysis methods that can be used to effectively extract the problem characteristics. Subsequently, we propose a classification of the landscape analysis methods based on their order, neighborhood structure and computational complexity. We then discuss applications of the algorithm selection framework and the relationship between it and algorithm portfolios, hybrid meta-heuristics, and hyper-heuristics. The paper concludes with the identification of key challenges and proposes future research directions.

Searching for structural bias in particle swarm optimization and differential evolution algorithms

“…Although in this study we pay attention mostly to the problem of structural bias in the algorithms (Kononova et al 2015), it is useful to place our discussion in a broader context. First of all, the abundance of metaheuristics that are frequently developed without any theoretical justification leads to at least three undesired effects: (i) it is hard to properly choose the right method for a particular task (Muñoz et al 2015;Yuen and Zhang 2015); (ii) although performance of some widely used metaheuristics is promising, many others show little (if any) novelty and efficiency-instead they rather compete for popularity by using appealing nomenclature (Sörensen 2015); (iii) even though studies of the behaviour of metaheuristics do appear (Clerc and Kennedy 2002;Van den Bergh and Engelbrecht 2004;Auger and Doerr 2011;Liao et al 2013;Bonyadi and Michalewicz 2014;Cleghorn and Engelbrecht 2014;Hu et al 2014;Rada-Vilela et al 2014;Leonard et al 2015;Hu et al 2016), in the majority of papers a desire to propose yet another novel optimizer dominates over the willingness to gain deeper insight into the already available methods. This fact, combined with a lack of commonly accepted procedures to properly develop, study and compare metaheuristics [see for example the discussion in Michalewicz 2012 andSörensen et al 2015], triggered a number of critical research outputs that identified many important issues.…”

Analyzing adaptive parameter landscapes in parameter adaptation methods for differential evolution

“…An analysis of L f can provide useful information even for blackbox optimization. For example, if the features of a given problem (e.g., ruggedness and neutrality) becomes clear by analyzing L f , an appropriate optimizer can be selected [31]. A number of methods for analyzing L f have been proposed in the literature [27,35].…”

A Systematic Approach for the Selection of Optimization Algorithms including End-User Requirements Applied to Box-Type Boom Crane Design

“…Heuristics are problem-tailored algorithms and metaheuristics are general purpose optimization algorithms, often imitating natural phenomena [22,23]. The task of selecting the most appropriate algorithm from the class of direct search methods remains a difficult challenge [24].…”