Investigations into Lamarckism; Baldwinism and local search in grammatical evolution guided by reinforcement Articles uri icon

authors

publication date

  • January 2013

start page

  • 595

end page

  • 627

issue

  • 3

volume

  • 32

International Standard Serial Number (ISSN)

  • 1335-9150

abstract

  • Grammatical Evolution Guided by Reinforcement is an extension of Grammatical Evolution that tries to improve the evolutionary process adding a learning process for all the individuals in the population. With this aim, each individual is given a chance to learn through a reinforcement learning mechanism during its lifetime. The learning process is completed with a Lamarckian mechanism in which an original genotype is replaced by the best learnt genotype for the individual. In a way, Grammatical Evolution Guided by Reinforcement shares an important feature with other hybrid algorithms, i.e. global search in the evolutionary process combined with local search in the learning process. In this paper the role of the Lamarck Hypothesis is reviewed and a solution inspired only in the Baldwin effect is included as well. Besides, different techniques about the trade-off between exploitation and exploration in the reinforcement learning step followed by Grammatical Evolution Guided by Reinforcement are studied. In order to evaluate the results, the system is applied on two different domains: a simple autonomous navigation problem in a simulated Kephera robot and a typical Boolean function problem.

keywords

  • computer science; artificial intelligence; hybrid algorithms; lamarckism; grammatical evolution; baldwinism