Efficacy of modified backpropagation and optimisation methods on a real-world medical problem

D. Alpsan, M. Towsey, Ozcan Ozdamar, A. C. Tsoi, D. N. Ghista

Research output: Contribution to journal › Article

32 Citations (Scopus)

Abstract

A wide range of modifications to the backpropagation (BP) algorithm, motivated by heuristic arguments and optimisation theory, has been examined on a real-world medical signal classification problem. The method of choice depends both upon the nature of the learning task and whether one wants to optimise learning for speed or generalisation. It was found that, comparatively, standard BP was sufficiently fast and provided good generalisation when the task was to learn the training set within a given error tolerance. However, if the task was to find the global minimum, then standard BP failed to do so within 100000 iterations, but first order methods which adapt the stepsize were as fast as, if not faster than, conjugate gradient and quasi-Newton methods. Second order methods required the same amount of fine tuning of line search and restart parameters as did the first order methods of their parameters in order to achieve optimum performance.

Original languageEnglish
Pages (from-to)945-962
Number of pages18
JournalNeural Networks
Volume8
Issue number6
DOIs
StatePublished - Jan 1 1995

Fingerprint

Medical problems
Backpropagation
Backpropagation algorithms
Newton-Raphson method
Tuning
Learning
Generalization (Psychology)

Keywords

  • Auditory evoked potential
  • Backpropagation
  • Multilayer perception
  • Neural networks
  • Optimisation
  • Pattern classification

ASJC Scopus subject areas

  • Artificial Intelligence
  • Neuroscience(all)
  • Cognitive Neuroscience

Cite this

Efficacy of modified backpropagation and optimisation methods on a real-world medical problem. / Alpsan, D.; Towsey, M.; Ozdamar, Ozcan; Tsoi, A. C.; Ghista, D. N.

In: Neural Networks, Vol. 8, No. 6, 01.01.1995, p. 945-962.

Research output: Contribution to journal › Article

Alpsan, D, Towsey, M, Ozdamar, O, Tsoi, AC & Ghista, DN 1995, 'Efficacy of modified backpropagation and optimisation methods on a real-world medical problem', Neural Networks, vol. 8, no. 6, pp. 945-962. https://doi.org/10.1016/0893-6080(95)00034-W
Alpsan D, Towsey M, Ozdamar O, Tsoi AC, Ghista DN. Efficacy of modified backpropagation and optimisation methods on a real-world medical problem. Neural Networks. 1995 Jan 1;8(6):945-962. https://doi.org/10.1016/0893-6080(95)00034-W
Alpsan, D. ; Towsey, M. ; Ozdamar, Ozcan ; Tsoi, A. C. ; Ghista, D. N. / Efficacy of modified backpropagation and optimisation methods on a real-world medical problem. In: Neural Networks. 1995 ; Vol. 8, No. 6. pp. 945-962.
@article{86714f95ed4d4ef8afd8402c2c9b440d,
title = "Efficacy of modified backpropagation and optimisation methods on a real-world medical problem",
abstract = "A wide range of modifications to the backpropagation (BP) algorithm, motivated by heuristic arguments and optimisation theory, has been examined on a real-world medical signal classification problem. The method of choice depends both upon the nature of the learning task and whether one wants to optimise learning for speed or generalisation. It was found that, comparatively, standard BP was sufficiently fast and provided good generalisation when the task was to learn the training set within a given error tolerance. However, if the task was to find the global minimum, then standard BP failed to do so within 100000 iterations, but first order methods which adapt the stepsize were as fast as, if not faster than, conjugate gradient and quasi-Newton methods. Second order methods required the same amount of fine tuning of line search and restart parameters as did the first order methods of their parameters in order to achieve optimum performance.",
keywords = "Auditory evoked potential, Backpropagation, Multilayer perception, Neural networks, Optimisation, Pattern classification",
author = "D. Alpsan and M. Towsey and Ozcan Ozdamar and Tsoi, {A. C.} and Ghista, {D. N.}",
year = "1995",
month = "1",
day = "1",
doi = "10.1016/0893-6080(95)00034-W",
language = "English",
volume = "8",
pages = "945--962",
journal = "Neural Networks",
issn = "0893-6080",
publisher = "Elsevier Limited",
number = "6",

}

TY - JOUR

T1 - Efficacy of modified backpropagation and optimisation methods on a real-world medical problem

AU - Alpsan, D.

AU - Towsey, M.

AU - Ozdamar, Ozcan

AU - Tsoi, A. C.

AU - Ghista, D. N.

PY - 1995/1/1

Y1 - 1995/1/1

N2 - A wide range of modifications to the backpropagation (BP) algorithm, motivated by heuristic arguments and optimisation theory, has been examined on a real-world medical signal classification problem. The method of choice depends both upon the nature of the learning task and whether one wants to optimise learning for speed or generalisation. It was found that, comparatively, standard BP was sufficiently fast and provided good generalisation when the task was to learn the training set within a given error tolerance. However, if the task was to find the global minimum, then standard BP failed to do so within 100000 iterations, but first order methods which adapt the stepsize were as fast as, if not faster than, conjugate gradient and quasi-Newton methods. Second order methods required the same amount of fine tuning of line search and restart parameters as did the first order methods of their parameters in order to achieve optimum performance.

AB - A wide range of modifications to the backpropagation (BP) algorithm, motivated by heuristic arguments and optimisation theory, has been examined on a real-world medical signal classification problem. The method of choice depends both upon the nature of the learning task and whether one wants to optimise learning for speed or generalisation. It was found that, comparatively, standard BP was sufficiently fast and provided good generalisation when the task was to learn the training set within a given error tolerance. However, if the task was to find the global minimum, then standard BP failed to do so within 100000 iterations, but first order methods which adapt the stepsize were as fast as, if not faster than, conjugate gradient and quasi-Newton methods. Second order methods required the same amount of fine tuning of line search and restart parameters as did the first order methods of their parameters in order to achieve optimum performance.

KW - Auditory evoked potential

KW - Backpropagation

KW - Multilayer perception

KW - Neural networks

KW - Optimisation

KW - Pattern classification

UR - http://www.scopus.com/inward/record.url?scp=0028791125&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=0028791125&partnerID=8YFLogxK

U2 - 10.1016/0893-6080(95)00034-W

DO - 10.1016/0893-6080(95)00034-W

M3 - Article

AN - SCOPUS:0028791125

VL - 8

SP - 945

EP - 962

JO - Neural Networks

JF - Neural Networks

SN - 0893-6080

IS - 6

ER -

Access to Document