Jove
Visualize
Contact Us

Related Experiment Videos

Optimization of the kernel functions in a probabilistic neural network analyzing the local pattern distribution.

I Galleske1, J Castellanos

  • 1Departmento de Inteligencia Artificial, Facultad de Informática, Universidad Politécnica de Madrid, Spain. gingo@zipi.fi.upm.es

Neural Computation
|April 26, 2002
PubMed
Summary

This study introduces an automated method for determining the covariance matrix in probabilistic neural networks (PNNs). This approach enhances network generalization and classification accuracy, outperforming standard PNNs and other 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

EMC Effect of Tritium and Helium-3 from the JLab MARATHON Experiment.

Physical review letters·2025
Same author

Novel Measurement of the Neutron Magnetic Form Factor from A=3 Mirror Nuclei.

Physical review letters·2024
Same author

Revealing the short-range structure of the mirror nuclei <sup>3</sup>H and <sup>3</sup>He.

Nature·2022
Same author

Measurement of the Nucleon F_{2}^{n}/F_{2}^{p} Structure Function Ratio by the Jefferson Lab MARATHON Tritium/Helium-3 Deep Inelastic Scattering Experiment.

Physical review letters·2022
Same author

Form Factors and Two-Photon Exchange in High-Energy Elastic Electron-Proton Scattering.

Physical review letters·2022
Same author

Probing Few-Body Nuclear Dynamics via ^{3}H and ^{3}He (e,e^{'}p)pn Cross-Section Measurements.

Physical review letters·2020
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

Area of Science:

  • Machine Learning
  • Artificial Intelligence
  • Pattern Recognition

Background:

  • Probabilistic Neural Networks (PNNs) rely on the Gaussian kernel function, whose covariance matrix is crucial for performance.
  • Manual determination of the covariance matrix is complex and limits network optimization.
  • Existing PNN models often struggle with generalization and classification accuracy on complex datasets.

Purpose of the Study:

  • To propose an automated procedure for determining the covariance matrix elements of the Gaussian kernel function in PNNs.
  • To enhance the generalization ability and classification performance of PNNs.
  • To provide a method that frees designers from manual covariance matrix specification.

Main Methods:

  • Automatic determination of rotation and variance matrices by analyzing the local environment of training patterns.

Related Experiment Videos

  • Combining these matrices to form the covariance matrix for each training pattern.
  • Validation using a variation of the two-spiral problem and UCI Machine Learning Repository datasets.

    • Main Results:

    • The automated method successfully determined covariance matrix elements.
    • The proposed PNN model demonstrated superior classification rates compared to the original PNN.
    • The model outperformed other well-known classification techniques on tested datasets.

    Conclusions:

    • Automated covariance matrix determination significantly improves PNN performance and generalization.
    • This method offers a practical advantage by eliminating manual parameter tuning.
    • The approach is effective for both synthetic and real-world classification tasks.