Source:
Evolutionary Programming V: Proceedings of the Fifth Annual Conference on Evolutionary Programming, MIT Press, San Diego (1996)
ISBN:
0-262-06190-2
URL:
http://www.genetic-programming.com/jkpdf/ep1996.pdf
Keywords:
genetic algorithms;
genetic programming
Abstract:
The solution to many problems requires, or is
facilitated by, the use of iteration. Moreover, because
iterative steps are repeatedly executed, they must have
some degree of generality. An automatic programming
system should require that the user make as few
problem-specific decisions as possible concerning the
size, shape, and character of the ultimate solution to
the problem. Work first presented at the Fourth Annual
Conference on Evolutionary Programming in 1995 (EP-95)
demonstrated that six then-new architecture-altering
operations made it possible to automate the decision
about the architecture of an overall program
dynamically during a run of genetic programming. The
question arises as to whether it is also possible to
automate the decision about whether to employ
iteration, how much iteration to employ, and the
particular sequence of iterative steps. This paper
introduces the new operation of restricted iteration
creation that automatically creates a restricted
iteration-performing branch out of a portion of an
existing computer program during a run a genetic
programming. Genetic programming with the new operation
is then used (in conjunction with the other
architecture-altering operations first presented at
EP-95) to evolve a computer program to solve a
non-trivial problem