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

Using Features for the Storage of Patterns in a Fully Connected Net.

J G. Taylor1, S Coombes

  • 1Centre for Neural Networks, Kings College, London, UK

Neural Networks : the Official Journal of the International Neural Network Society
|July 1, 1996
PubMed
Summary

This study introduces a feature matrix for Hopfield networks, enabling storage of correlated patterns using principal components. Simulations confirm its capacity to store N patterns in N spins, with a biologically plausible learning rule discussed.

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

Phase and amplitude responses for delay equations using harmonic balance.

Physical review. E·2024
Same author

Insights into oscillator network dynamics using a phase-isostable framework.

Chaos (Woodbury, N.Y.)·2024
Same author

Standing and travelling waves in a spherical brain model: The Nunez model revisited.

Physica D. Nonlinear phenomena·2017
Same author

Neuronal spike-train responses in the presence of threshold noise.

Frontiers in life science·2015
Same author

Unifying principles of calcium wave propagation - Insights from a three-dimensional model for atrial myocytes.

Biochimica et biophysica acta·2015
Same author

Pulsating fronts in periodically modulated neural field models.

Physical review. E, Statistical, nonlinear, and soft matter physics·2011

Area of Science:

  • Computational neuroscience
  • Artificial neural networks
  • Statistical mechanics

Background:

  • Hopfield networks are models for associative memory.
  • Storing correlated and biased patterns is a challenge in neural network models.
  • Existing methods like the pseudo-inverse rule have limitations.

Purpose of the Study:

  • To derive a novel set of weights (feature matrix) for storing correlated biased patterns in Hopfield networks.
  • To analyze the storage capacity and basins of attraction for this feature matrix.
  • To propose a biologically plausible learning rule for implementing the feature matrix.

Main Methods:

  • Derivation of network weights based on the criterion of local field vector as pattern reconstruction.
  • Utilizing eigenvectors/principal components of the pattern correlation matrix to define the feature matrix.

Related Experiment Videos

  • Employing network simulations to investigate storage capacity and basins of attraction.
  • Applying statistical mechanical analysis with the replica trick to confirm storage capacity.

    • Main Results:

    • The feature matrix demonstrates a storage capacity of up to N patterns in a network of N spins.
    • Simulation results for basins of attraction are presented and compared with theoretical analysis and the pseudo-inverse rule.
    • Statistical mechanical analysis validates the derived storage capacity.

    Conclusions:

    • The feature matrix offers an effective method for storing correlated biased patterns in Hopfield networks.
    • The proposed method achieves high storage capacity and is supported by theoretical and simulation evidence.
    • A biologically plausible learning rule for realizing the feature matrix in neural networks is discussed.