Building the Building Blocks: From Simplification to Winning Trees in Genetic Programming
Genetic Programming (GP) represents a powerful paradigm in diverse real-world applications. While GP can reach optimal (or at least ``good-enough'') solutions for many problems, such solutions are not without deficiencies. A frequent issue stems from the representation perspective where GP evolves solutions that contain unnecessary parts, known as program bloat. This paper first investigates a combination of deterministic and random simplification to simplify the solutions while having a (relatively) small influence on the solution fitness. Afterward, we use the solutions to extract their subtrees, which we denote as winning trees. The winning trees can be used to initialize the population for the new GP run and result in improved convergence and fitness, provided some conditions on the size of solutions and winning trees are fulfilled. To experimentally validate our approach, we consider several synthetic benchmark problems and real-world symbolic regression problems.
PDF Abstract
