abstract = "Optimisation within nearly infinite search space is a
common problem in applied science, for which two
examples illustrate the application of genetic
programming. Three different techniques were used to
develop filters for the removal of noise from
experimental data. Heuristic search was used to develop
a median filter, a classical genetic algorithm
optimized a 7-tap moving average (FIR) filter, and
genetic programming was used to optimize a stack
filter. The latter had the highest fitness, and was
computationally more efficient than the best median
filter, which was in turn superior in fitness to the
best moving average filter. Genetic programming was
also used to fit empirical equations to a chaotic
time-series (the Mackey-Glass equation) and non-linear
physiological data. Initial results confirm the key
role of the fitness measure in such work; oscillatory
series are readily fitted with linear functions unless
the computation of fitness includes an appropriate
measure such as incremental comparison of Fourier power
series. The use of Lyapunov exponents and dimension
estimation is suggested in more sophisticated compound
fitness measures. Genetic programming may prove to be
useful in both forecasting and structural studies of
non-linear systems, at both local and global levels.",
notes = "Two Scientific Applications of Genetic Programming:
The development of stack filters, the fitting of
non-linear equations to chaotic data
Mackey-Glass, REAL World examples. Contrasts with other
techniques eg GA.