In recent years biologists developed new interpretations and theories for understanding the mechanisms of evolution, but the biological foundations of evolutionary algorithms has not been debated with an analogous fervor. Most of the work performed by biologists has not been considered for inclusion into the Evolutionary Algorithm framework yet, and important debates between biologists have been almost completely disregarded. An interesting controversy that recently puzzled the biological world concerns natural selection.

The orthodox premise is that natural selection acts on the individual or, more properly, on differences among individuals within a population. However, in recent years, different authors argued that evolution can be better explained considering selection at higher levels such as kin, groups or species, while some others claimed that is preferable a lower level, such as genes.

The theory of gene selection was introduced in mid 70s by Richard Dawkins in his famous book, The Selfish Gene, and subsequently developed in The Extended Phenotype. In these works, Dawkins argues that genes, and genes alone, are the units of selection (replicators in his terminology) while organisms are simple vehicles, the packaging for replicators.

With the Selfish Gene Algorithm (SG), we propose to apply the shift of paradigm brought by the selfish gene theory to the field of Evolutionary Computation. We propose to rely directly on genes as unit of selection. In our opinion the simulated selection mechanism should be able to maintain essential gene correlation (i.e., the fact that a gene can be good or bad depending on the context of the other genes) as the real process does. In fact, in nature linkages between genes tend to emerge although, according to Dawkins' claims, selection works locally on genes.

The SG algorithm can leads to a very fast convergence towards a local optimum, but still explores the neighborhood of an upward sloping path. Furthermore the SG, adopting a mechanism similar to biological speciation, is able to discover interdependencies between genes even with highly deceptive, epistatic fitness functions like John Holland's Royal Roads.

Papers

Papers exploiting the Selfish Gene Algorithm published by the CAD Group:

Contact information

Giovanni Squillero

Politecnico di Torino Dipartimento di Automatica e Informatica Corso Duca degli Abruzzi 24 I-10129 Torino ITALY

Tel: +39-011564.7186 Fax: +39-011564.7099

E-mail: giovanni . squillero Personal web page: http://www.cad.polito.it/staff/squillero/ Politecnico info page: http://www.dauin.polito.it/en/personale/scheda/(matricola)/003584

© by CAD Group. Licensed under Creative Commons Attribution-Noncommercial-Share Alike 3.0 License. © by CAD Group. Licensed under Creative Commons Attribution-Noncommercial-Share Alike 3.0 License.