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

Dynamic proximity of spatio-temporal sequences.

David Horn1, Gideon Dror, Brigitte Quenet

  • 1School of Physics and Astronomy, Raymond and Beverly Sackler Faculty of Exact Sciences, Tel Aviv University, Tel Aviv 69978, Israel. horn@post.tau.ac.il

IEEE Transactions on Neural Networks
|October 16, 2004
PubMed
Summary

Comparing synaptic weight matrices reveals proximity between recurrent neural network sequences. Large datasets enable accurate matrix determination, enabling spatio-temporal sequences to encode bias vectors effectively.

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

A genome-wide genetic screen identifies a novel kDNA replication protein in trypanosomes.

Nucleic acids research·2026
Same author

Lossless resistive micro-heater design for reconfigurable phase-change photonics.

Optics letters·2026
Same author

FMOPhore for hotspot identification and efficient fragment-to-lead growth strategies.

Nature communications·2026
Same author

Genetic origins and proteomic consequences of kinetoplast loss in trypanosomes.

PLoS pathogens·2026
Same author

Acoziborole resistance associated mutations in Trypanosoma brucei CPSF3.

PLoS pathogens·2026
Same author

Decoding efficacy and resistance space at a drug binding site.

Nature communications·2026

Area of Science:

  • Computational Neuroscience
  • Machine Learning

Background:

  • Recurrent neural networks (RNNs) generate complex spatio-temporal sequences.
  • These sequences can appear random but are governed by underlying synaptic weight matrices.

Purpose of the Study:

  • To define a proximity measure between RNN-generated spatio-temporal sequences.
  • To explore the relationship between sequences and their generating synaptic matrices.

Main Methods:

  • Utilized the dynamic neural filter (DNF) formalism to compare teacher and student binary neuron RNNs.
  • Employed large training datasets exceeding the Cover limit for synaptic matrix determination.
  • Introduced a linear support vector machine (SVM) variant for optimal weight matrix specification and noise handling.

Related Experiment Videos

Main Results:

  • Demonstrated that large sequences allow for accurate determination of synaptic matrices.
  • Showed that known matrices enable rapid bias vector determination, treating sequences as encodings.
  • Established that correlations of reconstructed synaptic matrices can compare different DNFs.

Conclusions:

  • Synaptic weight matrices provide a method for quantifying proximity between RNN spatio-temporal sequences.
  • The DNF formalism and SVM variants offer robust tools for analyzing and comparing neural network dynamics.
  • Spatio-temporal sequences can be interpreted as encodings of bias vectors, offering insights into network function.