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

Intelligent initialization of resource allocating RBF networks.

Manolis Wallace1, Nicolas Tsapatsoulis, Stefanos Kollias

  • 1Department of Computer Science, University of Indianapolis, Athens Campus, Athens, Greece. wallace@image.ntua.gr

Neural Networks : the Official Journal of the International Neural Network Society
|March 30, 2005
PubMed
Summary

This study introduces a novel unsupervised clustering method for initializing neural networks, simplifying parameter setup for multi-dimensional data. The approach yields compact, efficient networks with superior classification performance without manual tuning.

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

An End-to-End Deep Learning Framework for Fault Detection in Marine Machinery.

Sensors (Basel, Switzerland)·2024
Same author

Insights into Object Semantics: Leveraging Transformer Networks for Advanced Image Captioning.

Sensors (Basel, Switzerland)·2024
Same author

Evaluating the use of Instagram images color histograms and hashtags sets for automatic image annotation.

Frontiers in big data·2023
Same author

VACS: VAccination disComfort Scale.

Clinics and practice·2022
Same author

Safe university: a guide for open academic institutions through the pandemic.

Clinical microbiology and infection : the official publication of the European Society of Clinical Microbiology and Infectious Diseases·2022
Same author

Opinion Mining From Social Media Short Texts: Does Collective Intelligence Beat Deep Learning?

Frontiers in robotics and AI·2021

Area of Science:

  • Computer Science
  • Artificial Intelligence
  • Machine Learning

Background:

  • Proper neural network parameter initialization is crucial for efficient training, model compactness, and generalization.
  • Initializing parameters for multi-dimensional data is challenging and labor-intensive.
  • Resource Allocating Networks (RAN) benefit significantly from effective initialization strategies.

Purpose of the Study:

  • To propose a novel, multi-dimensional, unsupervised clustering method for initializing neural network parameters.
  • To automate the initialization process for Resource Allocating Networks (RAN) without user intervention.
  • To achieve compact, efficient neural network architectures with improved classification accuracy.

Main Methods:

  • A novel multi-dimensional, unsupervised clustering technique was developed.

Related Experiment Videos

  • The clustering output directly determined both hidden and output layer parameters for RAN.
  • The method was applied to publicly available datasets including iris, Wisconsin, and ionosphere data.

    • Main Results:

    • The proposed method successfully initialized neural network parameters without manual input.
    • Resulting network structures were compact and efficient.
    • The initialized networks achieved optimal classification results on tested datasets.

    Conclusions:

    • Unsupervised clustering offers an effective solution for initializing neural network parameters, especially for multi-dimensional data.
    • The proposed method automates initialization, leading to compact, efficient, and high-performing neural networks.
    • This approach eliminates the need for manual parameter selection, streamlining the neural network design process.