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

A modified back-propagation method to avoid false local minima.

Yutaka Fukuoka1, Hideo Matsuki, Haruyuki Minamitani

  • 1Institute for Medical and Dental Engineering, Tokyo Medical and Dental University, Chiyoda-ku, Tokyo, Japan

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

This study introduces a modified back-propagation algorithm to prevent neural networks from converging to incorrect local minima. The new method adjusts network weights during training, improving learning accuracy and stability.

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

Application of correlated component analysis to dynamic PET time-activity curves denoising.

Annual International Conference of the IEEE Engineering in Medicine and Biology Society. IEEE Engineering in Medicine and Biology Society. Annual International Conference·2021
Same author

Tension Sensor Based on Fluorescence Resonance Energy Transfer Reveals Fiber Diameter-Dependent Mechanical Factors During Myelination.

Frontiers in cellular neuroscience·2021
Same author

[Evaluation of physical activity using smartphones and wearable devices in healthcare: Current situation and future perspectives].

[Nihon koshu eisei zasshi] Japanese journal of public health·2021
Same author

Improving contrast between gray and white matter of Logan graphical analysis' parametric images in positron emission tomography through least-squares cubic regression and principal component analysis.

Biomedical physics & engineering express·2021
Same author

Changes in Autonomic Nervous System and Axillary Temperature Caused by Electric Potential Therapy.

Annual International Conference of the IEEE Engineering in Medicine and Biology Society. IEEE Engineering in Medicine and Biology Society. Annual International Conference·2020
Same author

Transplanted Oligodendrocyte Progenitor Cells Survive in the Brain of a Rat Neonatal White Matter Injury Model but Less Mature in Comparison with the Normal Brain.

Cell transplantation·2020

Area of Science:

  • Artificial Intelligence
  • Machine Learning
  • Neural Networks

Background:

  • The standard back-propagation algorithm often suffers from slow learning and convergence to suboptimal local minima.
  • Local minima present a significant challenge in training artificial neural networks, hindering performance.

Purpose of the Study:

  • To address the problem of convergence to false local minima in back-propagation.
  • To propose a modified back-propagation method that enhances learning accuracy.

Main Methods:

  • A modified back-propagation technique is introduced.
  • Each connecting weight is multiplied by a factor within (0,1] at constant intervals during learning.
  • This approach aims to maintain a larger sigmoid derivative when error signals are substantial.

Related Experiment Videos

Main Results:

  • Numerical experiments were conducted to validate the proposed method.
  • The results demonstrated the effectiveness of the modified back-propagation in overcoming local minima issues.
  • The method shows promise for more reliable neural network training.

Conclusions:

  • The modified back-propagation method successfully addresses the issue of converging to false local minima.
  • This technique offers an improvement over the standard back-propagation algorithm for neural network training.
  • Further research can explore its application in complex deep learning architectures.