Session:

MOD - Modularity, regularity and hierarchy in open-ended evolutionary computation

Title:

Function Set Independent Genetic Programming

Authors:

John R. Woodward

Abstract:

The choice of function set can significantly affect the performance of a GP run, possibly due to the difference in complexity of a function when expressed in different function sets. If modules are used in the representation, the complexity of a solution is independent of the function set (up to a fixed constant). This result motivates a representation which can be searched independently of the function set. We represent modules as look up tables, providing the definition of the module. This avoids having to invent heuristics to identify modules. It also delays the implementation details of a module until a decomposition of the problem has been obtained. We present a hierarchical modular search algorithm which decomposes a problem and reduces the impact of the choice of function set.