AppGP: an alternative structural representation for GP

It has been shown that standard genetic programming using standard subtree crossover is prone to a form of structural convergence which makes it extremely difficult to make changes near the root, occasionally causing runs to become trapped in local maxima. Based on these structural limitations we propose a different tree representation, AppGP, which we hope will avoid this problem in some cases. In this paper, we describe this representation, and compare its performance to the performance of standard GP on a suite of test problems. We find that on all of the test problems, AppGP does no worse than standard GP, and in several it does considerably better, suggesting that the representation warrants further study.