Peter Whigham
Dept of Information Science, University of Otago
Many disciplines are interested in forming mathematical expressions
to
represent relationships in observed data. Standard statistical
techniques
offer approaches to developing equations, however they are often limited
in
their form of expression and the types of mathematical functions that
can be
used.
This seminar will describe an evolutionary search approach to discovering
equations
that can generalise a set of observed data. The talk will focus on
multivariate time
series problems from the domain of freshwater ecology. Searching for
several different
equation forms, including difference equations and the use of past
variables, will be
described. An example of searching within the constraints of
a process-based model
will also be given. An introduction to formal language theory
and genetic programming
will be presented as background to this talk.