Among the most promising and active research areas in heuristic optimisation is the field of adaptive memetic algorithms (AMAs). These gain much of their reported robustness by adapting the probability with which each of a set of local improvement operators is applied, according to an estimate of their current value to the search process. This paper addresses the issue of how the current value should be estimated. Assuming the estimate occurs over several applications of a meme, we consider whether the extreme or mean improvements should be used, and whether this aggregation should be global, or local to some part of the solution space. To investigate these issues, we use the well-established COMA framework that coevolves the specification of a population of memes (representing different local search algorithms) alongside a population of candidate solutions to the problem at hand. Two very different memetic algorithms are considered: the first using adaptive operator pursuit to adjust the probabilities of applying a fixed set of memes, and a second which applies genetic operators to dynamically adapt and create memes and their functional definitions. For the latter, especially on combinatorial problems, credit assignment mechanisms based on historical records, or on notions of landscape locality, will have limited application, and it is necessary to estimate the value of a meme via some form of sampling. The results on a set of binary encoded combinatorial problems show that both methods are very effective, and that for some problems it is necessary to use thousands of variables in order to tease apart the differences between different reward schemes. However, for both memetic algorithms, a significant pattern emerges that reward based on mean improvement is better than that based on extreme improvement. This contradicts recent findings from adapting the parameters of operators involved in global evolutionary search. The results also show that local reward schemes outperform global reward schemes in combinatorial spaces, unlike in continuous spaces. An analysis of evolving meme behaviour is used to explain these findings.

Parameter tuning of evolutionary algorithms (EAs) is attracting more and more interest. In particular, the sequential parameter optimization (SPO) framework for the model-assisted tuning of stochastic optimizers has resulted in established parameter tuning algorithms. In this paper, we enhance the SPO framework by introducing transformation steps before the response aggregation and before the actual modeling. Based on design-of-experiments techniques, we empirically analyze the effect of integrating different transformations. We show that in particular, a rank transformation of the responses provides significant improvements. A deeper analysis of the resulting models and additional experiments with adaptive procedures indicates that the rank and the Box-Cox transformation are able to improve the properties of the resultant distributions with respect to symmetry and normality of the residuals. Moreover, model-based effect plots document a higher discriminatory power obtained by the rank transformation.

In this paper we extend a previously proposed randomized landscape generator in combination with a comparative experimental methodology to study the behavior of continuous metaheuristic optimization algorithms. In particular, we generate two-dimensional landscapes with parameterized, linear ridge structure, and perform pairwise comparisons of algorithms to gain insight into what kind of problems are easy and difficult for one algorithm instance relative to another. We apply this methodology to investigate the specific issue of explicit dependency modeling in simple continuous estimation of distribution algorithms. Experimental results reveal specific examples of landscapes (with certain identifiable features) where dependency modeling is useful, harmful, or has little impact on mean algorithm performance. Heat maps are used to compare algorithm performance over a large number of landscape instances and algorithm trials. Finally, we perform a meta-search in the landscape parameter space to find landscapes which maximize the performance between algorithms. The results are related to some previous intuition about the behavior of these algorithms, but at the same time lead to new insights into the relationship between dependency modeling in EDAs and the structure of the problem landscape. The landscape generator and overall methodology are quite general and extendable and can be used to examine specific features of other algorithms.

Meta-modeling has become a crucial tool in solving expensive optimization problems. Much of the work in the past has focused on finding a good regression method to model the fitness function. Examples include classical linear regression, splines, neural networks, Kriging and support vector regression. This paper specifically draws attention to the fact that assessing model accuracy is a crucial aspect in the meta-modeling framework. Resampling strategies such as cross-validation, subsampling, bootstrapping, and nested resampling are prominent methods for model validation and are systematically discussed with respect to possible pitfalls, shortcomings, and specific features. A survey of meta-modeling techniques within evolutionary optimization is provided. In addition, practical examples illustrating some of the pitfalls associated with model selection and performance assessment are presented. Finally, recommendations are given for choosing a model validation technique for a particular setting.

Recent years have witnessed the great success of hyper-heuristics applying to numerous real-world applications. Hyper-heuristics raise the generality of search methodologies by manipulating a set of low level heuristics (LLHs) to solve problems, and aim to automate the algorithm design process. However, those LLHs are usually parameterized, which may contradict the domain independent motivation of hyper-heuristics. In this paper, we show how to automatically maintain low level parameters (LLPs) using a hyper-heuristic with LLP adaptation (AD-HH), and exemplify the feasibility of AD-HH by adaptively maintaining the LLPs for two hyper-heuristic models. Furthermore, aiming at tackling the search space expansion due to the LLP adaptation, we apply a heuristic space reduction (SAR) mechanism to improve the AD-HH framework. The integration of the LLP adaptation and the SAR mechanism is able to explore the heuristic space more effectively and efficiently. To evaluate the performance of the proposed algorithms, we choose the p -median problem as a case study. The empirical results show that with the adaptation of the LLPs and the SAR mechanism, the proposed algorithms are able to achieve competitive results over the three heterogeneous classes of benchmark instances.