The Relationship between Semantic Distance and Performance in Dynamic Symbolic Regression Problems

Created by W.Langdon from gp-bibliography.bib Revision:1.8120

@InProceedings{tuite:mendel:2014,

• author = "Cliodhna Tuite and Michael O'Neill and Anthony Brabazon",
• title = "The Relationship between Semantic Distance and Performance in Dynamic Symbolic Regression Problems",
• booktitle = "Mendel 2014 The 20th International Conference on Soft Computing",
• year = "2014",
• address = "Brno, Czech Republic",
• keywords = "genetic algorithms, genetic programming, dynamic environments, symbolic regression, semantic distance",
• URL = "http://ncra.ucd.ie/papers/mendel2014_tuite.pdf",
• abstract = "Several methods which apply genetic programming (GP) in dynamic optimisation environments implicitly assume that the smaller the semantic change in the optimization goal, the better the adaptation of the GP population to the new target. However, GP searches over genetic operator-based fitness landscapes. As such, relative distances between solution points in genetic operator-based landscapes may not be related to the semantic distances between points. Our experiments examine whether decreasing the semantic distance between first and second-period target functions in symbolic regression problems result in improved performance in the second period. As a control, we also investigate how re-initialising the GP population in the second period performs in comparison with using a continuous GP population across the two periods. We find that decreasing the semantic distance does result in better performance in the second period, and that re-initializing the GP population under performs a continuous population at low semantic distances.",
• notes = "http://www.mendel-conference.org/tmp/ScheduleMendel2014e.pdf",

}

Genetic Programming entries for Cliodhna Tuite Michael O'Neill Anthony Brabazon

Citations