Jove
Visualize
Contact Us
JoVE
x logofacebook logolinkedin logoyoutube logo
ABOUT JoVE
OverviewLeadershipBlogJoVE Help Center
AUTHORS
Publishing ProcessEditorial BoardScope & PoliciesPeer ReviewFAQSubmit
LIBRARIANS
TestimonialsSubscriptionsAccessResourcesLibrary Advisory BoardFAQ
RESEARCH
JoVE JournalMethods CollectionsJoVE Encyclopedia of ExperimentsArchive
EDUCATION
JoVE CoreJoVE BusinessJoVE Science EducationJoVE Lab ManualFaculty Resource CenterFaculty Site
Terms & Conditions of Use
Privacy Policy
Policies

Related Experiment Videos

Accuracy/diversity and ensemble MLP classifier design.

Terry Windeatt1

  • 1Center for Vision, Speech, and Signal Processing (CVSSP), School of Electronics and Physical Sciences, University of Surrey, Guildford, Surrey GU2 7XH, UK. t.windeatt@surrey.ac.uk

IEEE Transactions on Neural Networks
|September 28, 2006
PubMed
Summary
This summary is machine-generated.

Related Concept Videos

You might also read

Related Articles

Articles linked to this work by shared authors, journal, and citation graph.

Sort by
Same author

Ensemble pruning using spectral coefficients.

IEEE transactions on neural networks and learning systems·2014
Same author

Minimising added classification error using Walsh coefficients.

IEEE transactions on neural networks·2011
Same author

Embedded feature ranking for ensemble MLP classifiers.

IEEE transactions on neural networks·2011
See all related articles

This study introduces a novel measure to predict optimal training epochs for multilayer perceptrons (MLP) ensembles. The method, based on spectral representation, improves classifier performance and aids in tuning complex models.

Area of Science:

  • Machine Learning
  • Artificial Intelligence
  • Pattern Recognition

Background:

  • Tuning multilayer perceptrons (MLP) classifiers presents significant challenges.
  • Ensemble methods improve classification but require careful parameter optimization.
  • Predicting optimal training epochs is crucial for achieving peak performance.

Purpose of the Study:

  • To introduce a novel measure for predicting the optimal number of training epochs for MLP classifier ensembles.
  • To incorporate accuracy and diversity into a single predictive measure.
  • To extend the applicability of the measure to multiclass problems.

Main Methods:

  • A measure based on the spectral representation of a Boolean function is computed between pairs of training data patterns.

Related Experiment Videos

  • This measure characterizes the mapping from classifier decisions to target labels.
  • The technique is extended to multiclass problems using output coding, comparing random and one-per-class codes.

    • Main Results:

    • The proposed measure shows strong correlation with base-classifier test error, enabling prediction of optimal training epochs.
    • While correlation with ensemble test error is weaker, the measure effectively predicts optimal ensemble training epochs.
    • For multiclass extension, random output coding outperforms one-per-class coding, even with well-tuned base classifiers.

    Conclusions:

    • The developed measure offers a valuable tool for optimizing MLP ensemble training, addressing key parameter tuning difficulties.
    • The spectral representation-based measure provides insights into classifier behavior and aids in performance prediction.
    • Output coding strategies significantly impact multiclass performance, with random codes showing superior results.