Evolutionary methods for tuning a robot sound recognition system
View/ Open
stirling dissertation.pdf (2.426Mb)
Date
2006Item status
Restricted AccessAuthor
Stirling, Timothy
Metadata
Abstract
A spatially-dispersed GA with co-evolutionary methodology was developed to
artificially evolve temporal-parameters for a spiking neural-model of the cricket auditory system
capable of performing phonotaxis. Male chromosomes containing genes that encode for the
temporal properties of calling songs were simultaneously evolved in the co-evolutionary model.
The application of A.I. modelling to the evolution of cricket species and their mating behaviour is
reviewed. The GA model produced discrete spatial groupings of individuals, which had distinct
genetic code within the male and female chromosomes. Networks with neural-parameters set
by the female chromosome’s genes showed a higher phonotactic performance when responding
to songs produced by males within that group than to songs produced by males from other
groups, supporting conspecific preference of calling song. However, this effect varied greatly
between groups and trials. The algorithm’s behaviour is complex, dynamic and chaotic, with
highly dimensional data necessitating complex analysis. The resulting analysis does not provide
a clear or concise synopsis of the behaviour and has left some open questions that would require
further research.