Abstract
The use of variable-length genomes in evolutionary computation has applications in optimisation when the size of the search space is unknown, and provides a unique environment to study the evolutionary dynamics of genome structure. Here, we revisit crossover for linear genomes of variable length, identifying two crucial attributes of successful recombination algorithms: the ability to retain homologous structure, and to reshuffle variant information. We introduce direct measures of these properties—homology score and linkage score—and use them to review existing crossover algorithms, as well as two novel ones. In addition, we measure the performance of these crossover methods on three different benchmark problems, and find that variable-length genomes out-perform fixed-length variants in all three cases. Our homology and linkage scores successfully explain the difference in performance between different crossover methods, providing a simple and insightful framework for crossover in a variable-length setting.
Highlights
Evolutionary algorithms are a family of computational methods that utilise natural selection for global optimisation on a wide range of problem types
Our work presents a new look at crossover in variable-length linear genomes, highlighting the dual importance of correctly exchanging homologous information, and recombining unique variations in the two parents
We quantify these two goals by defining the homology and linkage score, which can be measured for any crossover operator and sequence divergence, and show that these two factors explain the difference between crossover methods
Summary
Evolutionary algorithms are a family of computational methods that utilise natural selection for global optimisation on a wide range of problem types. The connection between evolution in computation and nature has been a positive influence on both fields. Evolutionary computation has benefited from innovation inspired by insights in natural evolution since its inception. [1, 2] For evolutionary scientists, algorithms can function as in silico models, allowing specific aspects of evolution to be isolated and studied from a different perspective and with a level of control that is not possible in nature [3, 4]. Genes are not identified by position, but by context: signalling sequences allow a decentralised system
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a callDisclaimer: All third-party content on this website/platform is and will remain the property of their respective owners and is provided on "as is" basis without any warranties, express or implied. Use of third-party content does not indicate any affiliation, sponsorship with or endorsement by them. Any references to third-party content is to identify the corresponding services and shall be considered fair use under The CopyrightLaw.
