Session:

LBP - Late Breaking Papers

Title:

Search Operator Bias in Linearly Structured Genetic Programming

Authors:

Garnet C. Wilson
Malcolm I. Heywood

Abstract:

GA solutions to the job-shop scheduling problem demonstrate that significant amounts of code context exist. Such observations have led to the introduction of biased search operators. In this work, we recognize that similar conditions exist in linearly structured GP (L-GP). An empirical study is made when biased search operators are applied to the San Mateo Trail (strategy), Two Box (regression), and Liver Disease (classification) benchmark problems. A preference is observed for biased mutation alone in the case of the regression problem, whereas the strategy and classification problems appear to prefer the combination of both biased mutation and crossover.