Exploring the Application of GOMEA to Bit-string GE

posted Apr 12, 2018, 12:10 AM by Eric Medvet   [ updated Apr 12, 2018, 12:11 AM ]
  • ACM Genetic and Evolutionary Computation Conference (GECCO), 2018, Kyoto (Japan), to appear
  • Eric Medvet, Alberto Bartoli, Andrea De Lorenzo
We explore the application of GOMEA, a recent method for discovering and exploiting the model for a problem in the form of linkage, to Grammatical Evolution (GE). GE employs an indirect representation based on familiar bit-string genotypes and is applicable to any problem where the solutions may be described using a context-free grammar, which hence greatly favors its wide adoption. Being general purpose, the representation of GE raises the opportunity for benefiting from the potential of GOMEA to automatically discover and exploit the linkage. We analyze experimentally the application of GOMEA to two bit-string-based variants of GE representation (the original representation and the recent WHGE) and show that GOMEA is clearly beneficial when coupled to WHGE, whereas it delivers no significant advantages when coupled with GE.