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

Modelling with constructive backpropagation.

M Lehtokangas1

  • 1Tampere University of Technology, Signal Processing Laboratory, P.O. Box 553, FIN-33101, Tampere, Finland

Neural Networks : the Official Journal of the International Neural Network Society
|March 29, 2003
PubMed
Summary

Constructive Backpropagation (CBP) offers an efficient neural network structure learning method. This approach improves modeling capabilities for nonlinear processes, outperforming existing cascade-correlation techniques.

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

Fast initialization for cascade-correlation learning.

IEEE transactions on neural networks·2008
Same author

Modified cascade-correlation learning for classification.

IEEE transactions on neural networks·2008
Same author

Optimization of sigma amplitude threshold in sleep spindle detection.

Journal of sleep research·2001
Same author

Autoassociative MLP in sleep spindle detection.

Journal of medical systems·2000
Same author

Determining the number of centroids for CMLP network.

Neural networks : the official journal of the International Neural Network Society·2000
Same author

Feedforward neural network with adaptive reference pattern layer.

International journal of neural systems·1999

Area of Science:

  • Computational intelligence
  • Machine learning
  • Artificial neural networks

Background:

  • Neural networks are effective for nonlinear process modeling, but structure selection is critical.
  • Poor structure leads to underfitting, overfitting, or wasted resources.
  • Constructive methods, like cascade-correlation (CC) learning, build structure incrementally.

Purpose of the Study:

  • To propose and evaluate Constructive Backpropagation (CBP), a novel neural network structure learning technique.
  • To compare CBP's efficiency and performance against established methods like CC learning.
  • To explore CBP's flexibility in automatic structure adaptation, including unit addition and deletion.

Main Methods:

  • Developed Constructive Backpropagation (CBP), inspired by cascade-correlation learning.

Related Experiment Videos

  • Implemented error backpropagation through a single hidden layer.
  • Extended CBP for simultaneous unit addition and continuous structure adaptation (addition/deletion).

    • Main Results:

    • CBP demonstrates computational efficiency comparable to CC learning.
    • CBP offers simpler implementation and compatibility with stochastic optimization.
    • Time series modeling experiments show CBP significantly outperforms CC learning in modeling capabilities.

    Conclusions:

    • CBP provides an effective and flexible approach to neural network structure learning.
    • The method enhances modeling performance and offers advantages in implementation and adaptability.
    • CBP represents a valuable advancement for modeling complex nonlinear processes.