abstract = "We propose replacing the division operator used in
genetic programming with an analytic quotient
operator.We demonstrate that this analytic quotient
operator systematically yields lower mean squared
errors over a range of regression tasks, due
principally to removing the discontinuities or
singularities that can often result from using either
protected or unprotected division. Further, the
analytic quotient operator is differentiable. We also
show that the new analytic quotient operator stabilises
the variance of the intermediate quantities in the
tree.",
notes = "Discussion of NaN inf (IEEE754, 1985). Cites
\cite{keijzer03}. Analytic Quotient AQ(x,y) =
x/sqrt(1+y*y). Normal (single objective SOGP). MOGP
used to limit bloat (cf \cite{langdon:book} ). PCGP
doi:10.1162/106365602760234117 Tested on six regression
problems (do any of them have negative values?) 'using
the AQ operator always yielded the smallest mean test
errors' (sec 3). 'AQ yields the smallest standard
deviation' (except one case, sec 3). 'correlation
between low training error and low test error' (sec 4).
Fig 3 uses log scales. Bloat shown in Table 12.