Large scale continuous EDA using mutual information

Abstract

Where a licence is displayed above, please note the terms and conditions of the licence govern your use of this document. When citing, please reference the published version. Take down policy While the University of Birmingham exercises care and attention in making items available there are rare occasions when an item has been uploaded in error or has been deemed to be commercially or otherwise sensitive.

An eigenspace divide-and-conquer approach for large-scale optimization

“…It concurrently evolves all the variable subsets with different probabilistic models and shows superior performance over traditional EDAs and some other efficient algorithms on a set of benchmark functions. Xu et al [44] further improved EDA-MCC by replacing the linear correlation coefficient with mutual information such that the nonlinear dependencies among variables can be detected. Yang et al [18] proposed a self-evaluation evolution (SEE) algorithm by combing the DC strategy with the surrogate model technique.…”

“…In this subsection, the performance of EDC was assessed on 14 test functions from CEC'2005 test suite. As EDC employs GSM-GEDA as the optimizer, three EDA-based algorithms for large-scale optimization, including EDA-MCC [43], EDA-MCC-MI [44] and RP-EDA [49], were selected for comparison. And the canonical GSM-GEDA was also included in this experiment to serve as a baseline.…”

“…EDA-MCC, EDA-MCC-MI and RP-EDA have all been introduced in Section 2. To ensure the fairness of the comparison, we directly employed the source codes of EDA-MCC and EDA-MCC-MI provided by Xu et al [44] and the source code of RP-EDA provided by Kabán et al [49], and set their parameters to default values. As for GSM-GEDA, it was originally developed for solving low-and medium-dimensional problems.…”

A Review of Population-Based Metaheuristics for Large-Scale Black-Box Global Optimization—Part I

“…Dong et al [78] propose an EDA with model complexity control which applies a threshold to the global Pearson correlation matrix to find weakly correlated dimensions and models them with univariate distributions and partitions the remaining strongly correlated variables into a set of lower dimensional spaces each of which is modeled using a multivariate distribution. To alleviate the deficiencies of Pearson correlation with non-Gaussian samples, Xu et al [79] proposed to use mutual information instead to detect variable dependence. A major downside of using mutual information however is its high computational cost [22].…”

Bi-Hierarchical Cooperative Coevolution for Large Scale Global Optimization

“…Along this research direction, some researchers suggested taking all the variables as random ones and dividing them according to the correlation coefficient between each pair of them [25], [26]. Xu et al [27] indicated that the commonly-used Pearson correlation coefficient cannot properly depict the nonlinear variable interdependencies, and replaced it with mutual information. Recently, Yang et al [28] proposed a new decomposition algorithm named affinity propagation assisted and evolution consistency based decomposition (APEC).…”