 ...
\let

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
 ...^{2}
 Note that because size distributions are typically skewed,
the proportion of aboveaveragesize programs, , is not
necessarily 0.5. For example, in a population containing 5 programs
of size 1, 4 of size 2, 3 of size 3, 2 of size 4 and 1 of size 5,
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
 ...PoliEtAltab:successRates.^{3}
 While the results reported in
Table .2 provide a useful indication of
relative performance differences, they were obtained for one
particular definition of success: that the fitness is no greater
than 1. Naturally, different definitions of success may produce
different results.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
 ...^{4}
 In fact, preliminary attempts at solving the BCI problem
showed that we could not even perform full scale
runs without bloat control.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.