Selmar K. Smit

Parameter Tuning for Configuring and Analyzing Evolutionary Algorithms

In this paper we present a conceptual framework for parameter tuning, provide a survey of tuning methods, and discuss related methodological issues. The framework is based on a three-tier hierarchy of a problem, an evolutionary algorithm (EA), and a tuner. Furthermore, we distinguish problem instances, parameters, and EA performance measures as major factors, and discuss how tuning can be directed to algorithm performance and/or robustness. For the survey part we establish dierent taxonomies to categorize tuning methods and review existing work. Finally, we elaborate on how tuning can improve methodology by facilitating well-funded experimental comparisons and algorithm analysis.

Comparing parameter tuning methods for evolutionary algorithms

Tuning the parameters of an evolutionary algorithm (EA) to a given problem at hand is essential for good algorithm performance. Optimizing parameter values is, however, a non-trivial problem, beyond the limits of human problem solving.In this light it is odd that no parameter tuning algorithms are used widely in evolutionary computing. This paper is meant to be stepping stone towards a better practice by discussing the most important issues related to tuning EA parameters, describing a number of existing tuning methods, and presenting a modest experimental comparison among them. The paper is concluded by suggestions for future research - hopefully inspiring fellow researchers for further work.

Costs and Benefits of Tuning Parameters of Evolutionary Algorithms

We present an empirical study on the impact of different design choices on the performance of an evolutionary algorithm (EA). Four EA components are considered--parent selection, survivor selection, recombination and mutation--and for each component we study the impact of choosing the right operator, and of tuning its free parameter(s). We tune 120 different combinations of EA operators to 4 different classes of fitness landscapes, and measure the cost of tuning. We find that components differ greatly in importance. Typically the choice of operator for parent selection has the greatest impact, and mutation needs the most tuning. Regarding individual EAs however, the impact of design choices for one component depends on the choices for other components, as well as on the available amount of resources for tuning.

Parameter tuning of evolutionary algorithms: generalist vs. specialist

Finding appropriate parameter values for Evolutionary Algorithms (EAs) is one of the persistent challenges of Evolutionary Computing. In recent publications we showed how the REVAC (Relevance Estimation and VAlue Calibration) method is capable to find good EA parameter values for single problems. Here we demonstrate that REVAC can also tune an EA to a set of problems (a whole test suite). Hereby we obtain robust, rather than problem-tailored, parameter values and an EA that is a ‘generalist, rather than a ‘specialist. The optimized parameter values prove to be different from problem to problem and also different from the values of the generalist. Furthermore, we compare the robust parameter values optimized by REVAC with the supposedly robust conventional values and see great differences. This suggests that traditional settings might be far from optimal, even if they are meant to be robust.

Evolutionary Algorithm Parameters and Methods to Tune Them

In this chapter we discuss the notion of Evolutionary Algorithm (EAs) parameters and propose a distinction between EAs and EA instances, based on the type of parameters used to specify their details. Furthermore, we consider the most important aspects of the parameter tuning problem and give an overview of existing parameter tuning methods. Finally, we elaborate on the methodological issues involved here and provide recommendations for further development.

Beating the ‘world champion’ evolutionary algorithm via REVAC tuning

We present a case study demonstrating that using the REVAC parameter tuning method we can greatly improve the ‘world champion’ EA (the winner of the CEC-2005 competition) with little effort. For ‘normal’ EAs the margins for possible improvements are likely much bigger. Thus, the main message of this paper is that using REVAC great performance improvements are possible for many EAs at moderate costs. Our experiments also disclose the existence of ‘specialized generalists’, that is, EAs that are generally good on a set of test problems, but only w.r.t. one performance measure and not along another one. This shows that the notion of robust parameters is questionable and the issue requires further research. Finally, the results raise the question what the outcome of the CEC-2005 competition would have been, if all of EAs had been tuned by REVAC, but without further research it remains an open question whether we crowned the wrong king.

Using Entropy for Parameter Analysis of Evolutionary Algorithms

Evolutionary algorithms (EA) form a rich class of stochastic search methods that share the basic principles of incrementally improving the quality of a set of candidate solutions by means of variation and selection (Eiben and Smith 2003, De Jong 2006). Such variation and selection operators often require parameters to be specified. Finding a good set of parameter values is a nontrivial problem in itself. Furthermore, some EA parameters are more relevant than others in the sense that choosing different values for them affects EA performance more than for the other parameters. In this chapter we explain the notion of entropy and discuss how entropy can disclose important information about EA parameters, in particular, about their relevance. We describe an algorithm that is able to estimate the entropy of EA parameters and we present a case study, based on extensive experimentation, to demonstrate the usefulness of this approach and some interesting insights that are gained.

A generic approach to parameter control

On-line control of EA parameters is an approach to parameter setting that offers the advantage of values changing during the run. In this paper, we investigate parameter control from a generic and parameter-independent perspective. We propose a generic control mechanism that is targeted to repetitive applications, can be applied to any numeric parameter and is tailored to specific types of problems through an off-line calibration process. We present proof-of-concept experiments using this mechanism to control the mutation step size of an Evolutionary Strategy (ES). Results show that our method is viable and performs very well, compared to the tuning approach and traditional control methods.

It's fate: a self-organising evolutionary algorithm

We introduce a novel evolutionary algorithm where the centralized oracle ---the selection-reproduction loop--- is replaced by a distributed system of Fate Agents that autonomously perform the evolutionary operations. This results in a distributed, situated, and self-organizing EA, where candidate solutions and Fate Agents co-exist and co-evolve. Our motivation comes from evolutionary swarm robotics where candidate solutions evolve in real time and space. As a first proof-of-concept, however, here we test the algorithm with abstract function optimization problems. The results show that the Fate Agents EA is capable of evolving good solutions and it can cope with noise and changing fitness landscapes. Furthermore, an analysis of algorithm behavior also shows that this EA successfully regulates population sizes and adapts its parameters.

Predicting Individual Trip Destinations with Artificial Potential Fields

This paper presents a method to model the intended destination of a subject in real time, based on a trace of position information and prior knowledge of possible destinations. In contrast to most work in this field, it does so without the need for prior analysis of habitual travel patterns. The method models the certainty of each POI by means of a virtual charge, resulting in an artificial potential field that reflects the current estimate of the subject’s intentions. The virtual charges are updated as new information about the subject’s position arrives. We experimentally compare a number of update rules with various parameter settings, showing that it is important to take the distance to a potential destination into account when updating the charge.