Title

A Winner-Take-All Methodology: Finding The Best Evolutionary Algorithm For The Global Optimization Of Functions

Keywords

Differential Evolution; Particle Swarm Optimization; Single- objective optimization

Abstract

The problem of effectively and effciently finding the global optimum of a function by using evolutionary algorithms is current and pertinent, and two of the evolutionary techniques that have received significant attention in the literature are Particle Swarm Optimization (PSO), and Differential Evolution (DE), as well as their numerous variants. One way of taking advantage of the many good PSO and DE variant algorithms that have appeared in the literature is to run them all for a particular optimization problem and choose the best answer provided. This approach, referred to as the Naive Approach (NA) is time consuming. In this paper, we are using the naive approach with a suite of algorithms for each function minimization problem and we run the algorithms, in the suite, for a specifc number of function evaluations (typically much smaller than the number of function evaluations needed for each specific algorithm to converge to a solution) and decide, using appropriate performance measures of merit, which one of these algorithms will continue running until convergence; the rest of the algorithms, deemed as algorithms that will eventually produce inferior solutions, are aborted. We refer to our methodology, for obvious reasons, as the Winner-Take-All (WTA) methodology. Using this methodology we introduce WTA algorithmic variants that are effcient and effective solvers of global optimization problems. In this paper, we report results on one of these variants, called WTA1, and show that it is very competitive compared to the constituent algorithms in the suite used for its design, and effcient compared to NA. Copyright is held by the author/owner(s).

Publication Date

1-1-2012

Publication Title

GECCO'12 - Proceedings of the 14th International Conference on Genetic and Evolutionary Computation Companion

Number of Pages

1515-1516

Document Type

Article; Proceedings Paper

Personal Identifier

scopus

DOI Link

https://doi.org/10.1145/2330784.2331021

Socpus ID

84865039532 (Scopus)

Source API URL

https://api.elsevier.com/content/abstract/scopus_id/84865039532

This document is currently not available here.

Share

COinS