Peter A. Whigham
Lecturer, Department of Information Science
Title:
Evolution as Search
Abstract:
Darwin's concept of natural selection gave a convincing explanation
for the condition
of our natural environment. Elements from the theory of evolution may
also be used as
a basis for constructing machine-based systems for induction and optimisation.
This seminar will introduce the concept of evolution as search.
A minimal set of
the abstract components required to create a population-based search
will initially
be described. Using this framework the machine learning techniques
of Genetic Algorithms,
Genetic Programming, Evolving Neural Networks and Evolving Inductive
Logic Programs will
be introduced. Concepts such as fitness selection, mutation,
crossover, genotype,
phenotype, convergence, fitness landscapes, coevolution and epistasis
will be explained.
So are they actually useful? Come along and find out what the
answer is (and it's not
necessarily a resounding yes!). A number of applications will
be described which
demonstrate that they may have a role to play in model building and
system process
understanding, especially when those systems are complex or highly
non-linear.
This seminar is relevant to all researches interested in developing
theories for
time-series, spatial or qualitative data.