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

Fast computation with spikes in a recurrent neural network.

Dezhe Z Jin1, H Sebastian Seung

  • 1Howard Hughes Medical Institute and Department of Brain and Cognitive Sciences, Massachusetts Institute of Technology, Cambridge, Massachusetts 02139, USA. djin@mit.edu

Physical Review. E, Statistical, Nonlinear, and Soft Matter Physics
|June 13, 2002
PubMed
Summary

This study presents a neural network model that performs fast winner-take-all computations, challenging the notion that recurrent neural networks are too slow for brain modeling. The network rapidly identifies the neuron with the strongest input, even with varying initial states.

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

Uncovering Sex Differences in the Drosophila Ventral Nerve Cord Through Connectome Alignment.

bioRxiv : the preprint server for biology·2026
Same author

Eyewire II - A connectomic resource for resolving cell types and circuits of the mouse retina.

bioRxiv : the preprint server for biology·2026
Same author

Distributed control circuits across a brain-and-cord connectome.

Nature·2026
Same author

Temperature robustness of the timing network within songbird premotor nucleus HVC.

Journal of computational neuroscience·2026
Same author

Connectomic reconstruction from hippocampal CA3 reveals spatially graded mossy fiber inputs and selective feedforward inhibition to pyramidal cells.

bioRxiv : the preprint server for biology·2025
Same author

Distributed control circuits across a brain-and-cord connectome.

bioRxiv : the preprint server for biology·2025

Area of Science:

  • Computational Neuroscience
  • Artificial Neural Networks
  • Mathematical Biology

Background:

  • Recurrent neural networks are often considered computationally slow for modeling brain functions.
  • Integrate-and-fire neuron models are fundamental in computational neuroscience.
  • Winner-take-all (WTA) computations are crucial for decision-making and information processing in biological and artificial systems.

Purpose of the Study:

  • To analytically investigate a specific recurrent neural network architecture as a counterexample to the slowness assumption.
  • To demonstrate that such networks can perform rapid computations.
  • To characterize the conditions under which a winner-take-all computation is achieved.

Main Methods:

  • Analytical study of a network of N integrate-and-fire neurons.

Related Experiment Videos

  • Modeling self-excitation, all-to-all inhibition, and instantaneous synaptic coupling.
  • Analysis under constant external driving inputs and varying initial states.

    • Main Results:

    • The network performs a winner-take-all computation when inhibition and/or excitation are sufficiently large.
    • Computation is completed rapidly, often at the first spike of the winning neuron.
    • The number of potential winners (M) depends on the distribution of external inputs; initial states influence selection when M > 1.

    Conclusions:

    • Recurrent neural networks, specifically this integrate-and-fire model, can perform computations very quickly.
    • The network architecture facilitates efficient winner-take-all dynamics.
    • A specific balance of excitation and inhibition guarantees selection of the neuron with the maximal input.