Solution-Locked Averages and Solution-Time Binning in Genetic Programming


Poli R.

13th European Conference on Genetic Programming, İstanbul, Turkey, 7 - 09 April 2010, vol.6021, pp.208-219 identifier

  • Publication Type: Conference Paper / Full Text
  • Volume: 6021
  • City: İstanbul
  • Country: Turkey
  • Page Numbers: pp.208-219

Abstract

Averaging data collected in multiple independent runs across generations is the standard method to study the behaviour of GP. We show that while averaging may represent with good resolution GP's behaviour in the early stages of a run, it blurs later components of the dynamics. We propose two techniques to improve the situation: solution-locked averaging and solution-time binning. Results indicate that there are significant benefits in adopting these techniques.