Abstract
Anticorrelation has been used in training neural
network ensembles. Negative correlation learning (NCL)
is the state of the art anticorrelation measure. We
present an alternative anticorrelation measure,
RTQRTNCL, which shows significant improvements on our
test examples for both artificial neural networks (ANN)
and genetic programming (GP) learning machines. We
analyse the behaviour of the negative correlation
measure and derive a theoretical explanation of the
improved performance of RTQRTNCL in larger ensembles.
Users
Please
log in to take part in the discussion (add own reviews or comments).