Limiting code growth to improve robustness in tree-based genetic programming
Bart Wyns, Luc Boullart, Pieter Jan De Smedt · 2007
In this paper we analyze the composition of the function set ofthe artificial ant problem to define new training and testing trails similar to the Santa Fe trail. Cross-validation is used inapplications where large amounts of data are available. We alsouse a semantically driven growth limiter to reduce program sizeand check if growth reduction could lead to increased testperformance.