BPGEP is a very cool method- all the (relative) simplicity of PGEP, with the low arity representation of BGEP. It is by far the most fun I've had with randomly generated bit vectors, thats for sure.
So why not apply this simply representation, along with its nice, substructure preserving prefix notation, to other evolutionary techniques? Bit vectors are hardly uncommon, and the translation and interpretation stuff from bits to an expression tree is not very hard. Maybe I could do a binary particle swarm, or maybe even mix it up and do a generation of one technique, then another, and maybe some hill climbing in between.
The fact that we can go from bits to trees in such a nice way is very interesting to me, and I bet there is some cool stuff to be done with it. Operators that preserve substructure, combinations of evolutionary methods, the simplest possible example of random mutation hill climbing. Good stuff.
No comments:
Post a Comment