Using a genetic programming program to evolve a genetic programming
program has been thought of before. But, to my knowledge it has only
been applied to certain limited parameters (such as mutation rate and population size), not to the entire thing. By genetic programming
program, I mean the program that you wrote yourself, that mutates and
crosses over parse trees then assesses their fitness.
So, right now we are using our brains as the meta-genetic programmer.
We are using our brain to evolve (in a sense) the genetic programming program. But seeing as how the whole point of genetic programming is to imagine things that would not normally be thought up by the human
brain, why is it being done this way? If we can get a genetic
programming program to evolve on its own, we will take the limitations
of the brain's creativity out of the equation.
The process is simple in theory. Evolve a program that takes an
input-output table as the input, and outputs a program that
successfully runs this input output table as the output. This is after
all what your manually written genetic programming program does.
As we all know, the scalability problems of genetic programming would
make this impossible if it were started from a random seed. So, instead
it should be started from a population of genetic programming programs.
For example, the one you wrote yourself. Ones that other people have
written. Slight variations on each of those.
I would have already tried this out myself. But my intellect is
gradually beginning to fail me and I can only work at a snail's pace.
If this is a good idea I want to do it justice by testing it out as
soon as possible. I hope that one of you can help.
My Genetic Programming Web Site:
|Location:||Huddersfield, West Yorkshire, UK|
|Nodes:||8 (1 / 7)|