Stephen
Brooks / RAL / May 2004
Optimiser Architecture
•How
do you optimise in a very high-dimensional space?
–Hard
to calculate derivatives due to stochastic noise and sheer number of
dimensions
–Can
use a genetic algorithm
•Begins
with random designs
•Improves
with mutation, interpolation, crossover…
–Has
been highly successful so far in problems with up to 137 parameters