Hierarchical Self-Organization in Genetic programming

This paper presents an approach to automatic discovery of functions in Genetic Programming. The approach is based on discovery of useful building blocks by analyzing the evolution trace, generalizing blocks to define new functions, and finally adapting the problem representation onthe-fly. Adaptating the representation determines a hierarchical organization of the extended functionset whichenablesarestructuringofthesearch space so that solutions can be found more easily. Measures of complexity of solution trees are defined for an adaptive representation framework. The minimum description length principle is applied to justifythe feasibilityof approaches based on a hierarchy of discovered functionsand tosuggest alternative ways of defining a problem's fitness function. Preliminary empirical results are presented.

• Type: Conference
• Year: 1994
• Where: ICML
• Authors: Justinian P. Rosca, Dana H. Ballard
• Keywords: Adaptive Representation Framework, ICML 1994, Machine Learning, Problem Representation, Useful Building Blocks