Towards identifying populations that increase the likelihood of success in genetic programming

This paper presents a comprehensive, multivariate account of how initial population material is used over the course of a genetic programming run as while various factors influencing problem difficulty are changed. The results corroborate both theoretical and empirical studies on factors that influence population dynamics. The results also indicate a clue for a possible empirical measurement that could be used in tuning initial populations for increasing the likelihood of success. Categories and Subject Descriptors I.2.2 [Artificial Intelligence]: Automatic Programming – program synthesis. General Terms Algorithms, Performance, Experimentation, Theory. Keywords Population dynamics, initial populations, binomial-3, GP problem difficulty, building blocks, selection methods.

• Type: Conference
• Year: 2005
• Where: GECCO
• Authors: Jason M. Daida
• Keywords: GECCO 2005, Initial Populations, Population Dynamics, Problem Difficulty