Stochastic Diversity Loss and Scalability in Estimation of Distribution Genetic Programming

Created by W.Langdon from gp-bibliography.bib Revision:1.4448

  author =       "Kangil Kim and Bob (R. I.) Mckay",
  title =        "Stochastic Diversity Loss and Scalability in
                 Estimation of Distribution Genetic Programming",
  journal =      "IEEE Transactions on Evolutionary Computation",
  year =         "2013",
  volume =       "17",
  number =       "3",
  pages =        "301--320",
  month =        jun,
  keywords =     "genetic algorithms, genetic programming, Estimation of
                 Distribution Algorithm (EDA), Evolutionary Computation
                 (EC), Genetic Programming (GP), Likelihood Weighting
                 (LW), Probabilistic Prototype Tree (PPT), diversity
                 loss, sampling bias, sampling drift",
  ISSN =         "1089-778X",
  DOI =          "doi:10.1109/TEVC.2012.2196521",
  size =         "20 pages",
  abstract =     "In Estimation of Distribution Algorithms (EDA),
                 probability models hold accumulating evidence on the
                 location of an optimum. Stochastic sampling drift has
                 been heavily researched in EDA optimisation, but not in
                 EDAs applied to Genetic Programming (EDA-GP). We show
                 that, for EDA-GPs using Probabilistic Prototype Tree
                 (PPT) models, stochastic drift in sampling and
                 selection is a serious problem, inhibiting scaling to
                 complex problems. Problems requiring deep dependence in
                 their probability structure see such rapid stochastic
                 drift that the usual methods for controlling drift are
                 unable to compensate. We propose a new alternative,
                 analogous to likelihood weighting of evidence. We
                 demonstrate in a small-scale experiment that it does
                 counteract the drift, sufficiently to leave EDA-GP
                 systems subject to similar levels of stochastic drift
                 to other EDAs.",
  notes =        "Max problem \cite{langdon:1997:MAX} and onemax, 3 bit
                 even parity. Sampling drift, stochastic drift,
                 premature convergence. undefined allele U. 'The sample
                 size is reduced by the number of individuals sampled as
                 U.' 'The performance of both EDA systems is
                 substantially worse...' '2) Some probability tables
                 give zero probability to the correct alleles' 'They
                 won't help EDA-GP to scale to the problem complexities
                 typically handelled by today's GP systems...' imputing
                 missing values. porr performance on near trivial
                 problems DCTG-GP \cite{ross:2001:ngc} also known as

Genetic Programming entries for Kangil Kim R I (Bob) McKay