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

Neural net pruning based on functional behavior of neurons

N Shamir1, D Saad, E Marom

  • 1Faculty of Engineering, Tel Aviv University, Ramat Aviv, Israel.

International Journal of Neural Systems
|June 1, 1993
PubMed
Summary

This study introduces a novel neural network pruning technique that merges neurons based on functional behavior. This method reduces network complexity and enhances generalization by preserving functionality through a unique neuron merging process.

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

Artificial Intelligence Clustering Approach for Force Mapping Analysis of Polyacrylic Acid (PAA)/Polyethylene Oxide (PEO) Polymer Brushes for Biosensor Applications.

Journal of molecular recognition : JMR·2026
Same author

Communication networks beyond the capacity crunch.

Philosophical transactions. Series A, Mathematical, physical, and engineering sciences·2016
Same author

Disseminated histoplasmosis in allogeneic bone marrow transplant: a diagnosis not to be missed.

Transplant infectious disease : an official journal of the Transplantation Society·2014
Same author

High-efficiency arbitrary array generator.

Applied optics·2010
Same author

Two-dimensional wavelet transform achieved by computer-generated multireference matched filter and Dammann grating.

Applied optics·2010
Same author

Modified joint transform correlator binarized by error-diffusion. I. Spatially constant noise-dependent range limit.

Applied optics·2010

Area of Science:

  • Artificial Intelligence
  • Machine Learning
  • Deep Learning

Background:

  • Neural network pruning is crucial for reducing model complexity and improving generalization.
  • Existing methods often struggle to preserve network functionality after neuron removal.

Purpose of the Study:

  • To propose a novel pruning method based on neuron functional behavior.
  • To reduce neural network complexity and enhance generalization capabilities.
  • To preserve network functionality during the pruning process.

Main Methods:

  • Classifying neurons based on their internal representations across the training set.
  • Merging neurons with similar functional behaviors.
  • Implementing a unique merging and compensation procedure to transfer neuron roles.

Related Experiment Videos

Main Results:

  • Demonstrated implementation with a detailed numerical example.
  • Evaluated performance using statistical measures from repeated training.
  • Analyzed the influence of parameter selection on pruning and generalization.

Conclusions:

  • The proposed pruning method effectively reduces network complexity while preserving functionality.
  • Neuron functional behavior classification is a powerful tool for efficient network pruning.
  • The technique offers improved generalization capabilities for neural networks.