Correlation versus RMSE Loss Functions in Symbolic Regression Tasks

Created by W.Langdon from gp-bibliography.bib Revision:1.7615

@InProceedings{Banzhaf:2022:GPTP.2,

• author = "Nathan Haut and Wolfgang Banzhaf and Bill Punch",
• title = "Correlation versus {RMSE} Loss Functions in Symbolic Regression Tasks",
• booktitle = "Genetic Programming Theory and Practice XIX",
• year = "2022",
• editor = "Leonardo Trujillo and Stephan M. Winkler and Sara Silva and Wolfgang Banzhaf",
• series = "Genetic and Evolutionary Computation",
• pages = "31--55",
• address = "Ann Arbor, USA",
• month = jun # " 2-4",
• publisher = "Springer",
• keywords = "genetic algorithms, genetic programming",
• isbn13 = "978-981-19-8459-4",
• DOI = "doi:10.1007/978-981-19-8460-0_2",
• abstract = "The use of correlation as a fitness function is explored in symbolic regression tasks and its performance is compared against a more typical RMSE fitness function. Using correlation with an alignment step to conclude the evolution led to significant performance gains over RMSE as a fitness function. Employing correlation as a fitness function led to solutions being found in fewer generations compared to RMSE. We also found that fewer data points were needed in a training set to discover correct equations. The Feynman Symbolic Regression Benchmark as well as several other old and recent GP benchmark problems were used to evaluate performance.",
• notes = "Part of \cite{Banzhaf:2022:GPTP} published after the workshop in 2023",

}

Genetic Programming entries for Nathaniel Haut Wolfgang Banzhaf William F Punch

Citations