Evolvable Warps for Data Normalization

Created by W.Langdon from gp-bibliography.bib Revision:1.3872

@InProceedings{Gilbert:2016:CEC,
  author =       "Jeremy Gilbert and Daniel Ashlock",
  title =        "Evolvable Warps for Data Normalization",
  booktitle =    "Proceedings of 2016 IEEE Congress on Evolutionary
                 Computation (CEC 2016)",
  year =         "2016",
  editor =       "Yew-Soon Ong",
  pages =        "1562--1569",
  address =      "Vancouver",
  month =        "24-29 " # jul,
  publisher =    "IEEE Press",
  keywords =     "genetic algorithms, genetic programming",
  isbn13 =       "978-1-5090-0623-6",
  DOI =          "doi:10.1109/CEC.2016.7743975",
  abstract =     "The traditional method of fitting an approximate
                 cumulative probability distribution to a data set is to
                 bin the data in narrow bins and obtain a step function
                 approximation. This technique suffices for many
                 applications, but the resulting object is not a
                 differentiable function making recovery of the
                 underlying probability distribution function
                 impossible. In this study, a unique group theoretic
                 representation is used to define evolvable data warps
                 that can be used to recover continuous, infinitely
                 differentiable versions of the inverse cumulative
                 distribution function. The use of a group theoretic
                 representation permits a simple calculation to
                 transform the evolved object into a cumulative
                 distribution function and, via differentiation, into a
                 probability distribution function. The group used to
                 define the evolvable data warps is the group of
                 bijections of the unit interval. The generators used by
                 evolution are chosen to be differentiable in order to
                 enable the computation of probability distribution
                 functions. Experiments are run using a simple type of
                 evolutionary algorithm to evolve approximate CDFs on
                 seven data sets. The first data set is used to perform
                 a parameter study on the representation length used to
                 evolve the approximate CDFs and comparing two
                 variations of the representation - one of which uses a
                 representational control called gene expression and one
                 of which does not.",
  notes =        "WCCI2016",
}

Genetic Programming entries for Jeremy Gilbert Daniel Ashlock

Citations