Building predictive models in two stages with meta-learning templates optimized by genetic programming

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

@InProceedings{Kordik:2014:CIEL,
  author =       "Pavel Kordik and Jan Cerny",
  booktitle =    "IEEE Symposium on Computational Intelligence in
                 Ensemble Learning (CIEL 2014)",
  title =        "Building predictive models in two stages with
                 meta-learning templates optimized by genetic
                 programming",
  year =         "2014",
  month =        dec,
  keywords =     "genetic algorithms, genetic programming,
                 Meta-learning, Hierarchical ensembles, Ensemble
                 classifiers, Bagging, Boosting, Stacking, Cascade
                 generalisation, Non-stationary environments, Transfer
                 learning",
  DOI =          "doi:10.1109/CIEL.2014.7015740",
  size =         "8 pages",
  abstract =     "The model selection stage is one of the most difficult
                 in predictive modelling. To select a model with a
                 highest generalisation performance involves
                 benchmarking huge number of candidate models or
                 algorithms. Often, a final model is selected without
                 considering potentially high quality candidates just
                 because there are too many possibilities. Improper
                 benchmarking methodology often leads to biased
                 estimates of model generalisation performance.
                 Automation of the model selection stage is possible,
                 however the computational complexity is huge especially
                 when ensembles of models and optimisation of input
                 features should be also considered.

                 In this paper we show, how to automate model selection
                 process in a way that allows to search for complex
                 hierarchies of ensemble models while maintaining
                 computational tractability. We introduce two-stage
                 learning, meta-learning templates optimised by
                 evolutionary programming with anytime properties to be
                 able to deliver and maintain data-tailored algorithms
                 and models in a reasonable time without human
                 interaction.

                 Co-evolution if inputs together with optimisation of
                 templates enabled to solve algorithm selection problem
                 efficiently for variety of datasets.",
  notes =        "Also known as \cite{7015740}",
}

Genetic Programming entries for Pavel Kordik Jan Cerny

Citations