Towards unbiased benchmarking of evolutionary and hybrid algorithms for real-valued optimisation

MacNish, Cara

doi:10.1080/09540090701725581

Cited by 26 publications

(13 citation statements)

References 6 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…These are respectively known as Sphere, Schwefel's Problem 2.21, Rosenbrock's function, Rastrigin's function, Griewank's function and Ackley's function. The seventh is a fast fractal function, "DoubleDip", from the Fractal Function benchmarking suite [1], [9]. Functions 1, 4 and 6 are separable.…”

Section: Methodsmentioning

confidence: 99%

“…Many of the problems used to benchmark real-valued global optimisation algorithms are subject to biases [1], which may in turn give misleading results for the algorithms used to solve them. An example that has shown significant performance differences is directional bias, most commonly in the form of axial bias, where features of a problem surface are aligned with the Cartesian co-ordinate system in which it is defined.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Direction matters in high-dimensional optimisation

MacNish

Yao²

2008

2008 IEEE Congress on Evolutionary Computation (IEEE World Congress on Computational Intelligence)

Self Cite

View full text Add to dashboard Cite

Abstract-Directional biases are evident in many benchmarking problems for real-valued global optimisation, as well as many of the evolutionary and allied algorithms that have been proposed for solving them. It has been shown that directional biases make some kinds of problems easier to solve for similarly biased algorithms, which can give a misleading view of algorithm performance.In this paper we study the effects of directional bias for highdimensional optimisation problems. We show that the impact of directional bias is magnified as dimension increases, and can in some cases lead to differences in performance of many orders of magnitude.We present a new version of the classical evolutionary programming algorithm, which we call unbiased evolutionary programming (UEP), and show that it has markedly improved performance for high-dimensional optimisation.

show abstract

Section: Methodsmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Direction matters in high-dimensional optimisation

MacNish

Yao²

2008

2008 IEEE Congress on Evolutionary Computation (IEEE World Congress on Computational Intelligence)

Self Cite

View full text Add to dashboard Cite

show abstract

“…This includes the development of large-scale competitions and associated sets of benchmark test problems (e.g at recent Genetic and Evolutionary Computation Conferences (GECCO) and Congress on Evolutionary Computation (CEC)). Several different types of test problems have been used for the evaluation of algorithms, including constructed analytical functions, real-world problem instances or simplified versions of real-world problems and problem/landscape generators [3,4,5]. Different problem types have their own characteristics, however it is usually the case that complementary insights into algorithm behaviour result from conducting larger experimental studies using a variety of different problem types.…”

Section: Using a Landscape Generator To Actively Study The Relationshmentioning

confidence: 99%

When Does Dependency Modelling Help? Using a Randomized Landscape Generator to Compare Algorithms in Terms of Problem Structure

Morgan

Gallagher

2010

Parallel Problem Solving From Nature, PPSN XI

View full text Add to dashboard Cite

Abstract. In this paper we extend a previously proposed randomized landscape generator in combination with a comparative experimental methodology to study the behaviour of continuous metaheuristic optimization algorithms. In particular, we generate landscapes with parameterised, 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 modelling in simple continuous Estimation of Distribution Algorithms. Experimental results reveal specific examples of landscapes (with certain identifiable features) where dependency modelling is useful, harmful or has little impact on average algorithm performance. The results are related to some previous intuition about the behaviour of these algorithms, but at the same time lead to new insights into the relationship between dependency modelling in EDAs and the structure of the problem landscape. The overall methodology is quite general and could be used to examine specific features of other algorithms.

show abstract

“…We simulated the minimisation of Ackley's function, one of the common benchmarking functions given by Macnish [5], using a Gray coded and a real coded ga (in two dimensions for display purposes). Ackley's function is…”

Section: Simulation Examplesmentioning

confidence: 99%

Binary versus real coding for genetic algorithms: A false dichotomy?

Gaffney¹,

Pearce²

2010

ANZIAMJ

View full text Add to dashboard Cite

The usefulness of the genetic algorithm (ga) as judged by numerous applications in engineering and other contexts cannot be questioned. However, to make the application successful, often considerable effort is needed to customise the ga to suit the problem or class of problems under consideration. Perhaps the most basic decision which the designer of a ga makes, is whether to use binary or real coding. If the variable of the parameter space of an optimisation problem is continuous, a real coded ga is possibly indicated. Real numbers have a floating-point representation on a computer and the decision space is always discretised; it is not immediately evident that real coding should be the preferred method for encoding this particular problem. We re-visit this, and other decisions, which ga designers need to make. We present simulations on a standard test function, which show the result that no one ga performs best on every test problem. Perhaps the initial choice to code a problem using a real or binary coding is a false dichotomy. What counts are the algorithms for implementing

show abstract

Towards unbiased benchmarking of evolutionary and hybrid algorithms for real-valued optimisation

Cited by 26 publications

References 6 publications

Direction matters in high-dimensional optimisation

Direction matters in high-dimensional optimisation

When Does Dependency Modelling Help? Using a Randomized Landscape Generator to Compare Algorithms in Terms of Problem Structure

Binary versus real coding for genetic algorithms: A false dichotomy?

Contact Info

Product

Resources

About