Towards a method for automatic algorithm configuration: A design evaluation using tuners

Elizabeth Montero, María Cristina Riff

Research output: Contribution to journalArticlepeer-review

7 Citations (Scopus)

Abstract

Metaheuristic design is an incremental and difficult task. It is usually iterative and requires several evaluations of the code to obtain an algorithm with good performance. In this work, we analyse the design of metaheuristics by detecting components which are strictly necessary to obtain a good performance (in term of solutions quality). We use a collective strategy where the information generated by a tuner is used to detect the components usefulness. We evaluate this strategy with two well-known tuners EVOCA and I-RACE to analyse which one is more suitable and provides better results to make this components detection. The goal is to help the designer either to evaluate during the design process different options of the code or to simplify her/his final code without a loss in the quality of the solutions.

Original languageEnglish
Pages (from-to)90-99
Number of pages10
JournalLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume8672
Publication statusPublished - 1 Jan 2014

Keywords

  • Automated algorithm configuration
  • Automated algorithm tuning
  • Metaheuristics

ASJC Scopus subject areas

  • Theoretical Computer Science
  • Computer Science(all)

Fingerprint

Dive into the research topics of 'Towards a method for automatic algorithm configuration: A design evaluation using tuners'. Together they form a unique fingerprint.

Cite this