next up previous
Next: Acknowledgements Up:

Why Ants are Previous: Discussion


Conclusions

 

We have started an examination of the program space of a GP benchmark problem.

We have shown that there are many distinct solutions to the problem and the density of solutions in the program space is unexpectedly high. Indeed genetic programming and other search techniques do not perform enormously better than random search. Using the program landscape and schema analysis we have shown why the artificial ant following the Santa Fe trail problem is difficult for these search techniques and these suggest reasons why the Ant problem may be indicative of real problem spaces and so be worthy of further study.



William B Langdon
Tue Feb 3 19:53:29 GMT 1998