Complexity measures for multi-objective symbolic regression

Michael Kommenda, Andreas Beham, Michael Affenzeller, Gabriel Kronberger

Research output: Chapter in Book/Report/Conference proceedingsConference contributionpeer-review

12 Citations (Scopus)

Abstract

Multi-objective symbolic regression has the advantage that while the accuracy of the learned models is maximized, the complexity is automatically adapted and need not be specified a-priori. The result of the optimization is not a single solution anymore, but a whole Paretofront describing the trade-off between accuracy and complexity. In this contribution we study which complexity measures are most appropriately used in symbolic regression when performing multiobjective optimization with NSGA-II. Furthermore, we present a novel complexity measure that includes semantic information based on the function symbols occurring in the models and test its effects on several benchmark datasets. Results comparing multiple complexity measures are presented in terms of the achieved accuracy and model length to illustrate how the search direction of the algorithm is affected.

Original languageEnglish
Title of host publicationComputer Aided Systems Theory – EUROCAST 2015 - 15th International Conference, Revised Selected Papers
EditorsFranz Pichler, Roberto Moreno-Díaz, Alexis Quesada-Arencibia
PublisherSpringer
Pages409-416
Number of pages8
ISBN (Print)9783319273396
DOIs
Publication statusPublished - 2015
Event15th International Conference on Computer Aided Systems Theory, Eurocast 2015 - Las Palmas, Gran Canaria, Spain
Duration: 8 Feb 201513 Feb 2015
http://eurocast2015.fulp.ulpgc.es/

Publication series

NameLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume9520
ISSN (Print)0302-9743
ISSN (Electronic)1611-3349

Conference

Conference15th International Conference on Computer Aided Systems Theory, Eurocast 2015
Country/TerritorySpain
CityLas Palmas, Gran Canaria
Period08.02.201513.02.2015
Internet address

Keywords

  • Complexity measures
  • Genetic programming
  • Multiobjective optimization
  • NSGA-II
  • Symbolic regression

Fingerprint

Dive into the research topics of 'Complexity measures for multi-objective symbolic regression'. Together they form a unique fingerprint.

Cite this