Evolutionary Design of Neural Trees for Heart Rate Prediction

Created by W.Langdon from gp-bibliography.bib Revision:1.4216

  author =       "Byoung-Tak Zhang and Je-Gun Joung",
  title =        "Evolutionary Design of Neural Trees for Heart Rate
  booktitle =    "Soft Computing in Engineering Design and
  year =         "1997",
  editor =       "Pravir K. Chawdhry and Rajkumar Roy and Raj K. Pant",
  pages =        "93--101",
  publisher_address = "Godalming, GU7 3DJ, UK",
  month =        "23-27 " # jun,
  publisher =    "Springer-Verlag London",
  keywords =     "genetic algorithms, genetic programming, ANN",
  ISBN =         "3-540-76214-0",
  URL =          "http://www.springer.com/engineering/mechanical+eng/book/978-3-540-76214-0?cm_mmc=Google-_-Book%20Search-_-Springer-_-0",
  DOI =          "doi:10.1007/978-1-4471-0427-8_11",
  abstract =     "Some classes of neural networks are known as universal
                 function approximators. However, their training
                 efficiency and generalisation performance depend highly
                 on the structure which is usually determined by a human
                 designer. In this paper we present an evolutionary
                 computation method for automating the neural network
                 design process. We represent networks as tree
                 structures, called neural trees, in genotype and apply
                 genetic operators to evolve problem-dependent network
                 structures and their weights. Experimental results are
                 provided on the prediction of a heart rate time-series
                 by evolving sigma-pi neural trees.",
  notes =        "WSC2 Second On-line World Conference on Soft Computing
                 in Engineering Design and Manufacturing. feed forward
                 artificial neural networks comprised of sigma and pi
                 nodes evolved using GP. 'Between generations the
                 network weights are adapted by a stochastic
                 hill-climbing search.' Fitness based on error between
                 prediction and measurement on training examples plus
                 term related to complexity (ie size) of network and the
                 complexity of the best network in the previous
                 generation (ie fitness prefers parsimony). 'Without
                 (the parsimony term) the network size usually grows
                 without bound'.

                 wsc2/ind_paper/d_zhang.html URL broken 2005",
  size =         "9 pages",

Genetic Programming entries for Byoung-Tak Zhang Je-Gun Joung