Jove
Visualize
Contact Us

Related Experiment Videos

Models of MT and MST areas using wake-sleep algorithm.

Katsuki Katayama1, Masataka Ando, Tsuyoshi Horiguchi

  • 1Research Institute of Electrical Communication, Tohoku University, Sendai 980-8577, Japan.

Neural Networks : the Official Journal of the International Neural Network Society
|March 24, 2004
PubMed
Summary

We developed neural network models simulating middle temporal (MT) and medial superior temporal (MST) neurons. Models with more states (Q>=3) showed improved responses to optical flow compared to binary models.

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

Clinical Implementation Gap of Comprehensive Genomic Profiling in Pancreatic Cancer: A Real-world Study.

In vivo (Athens, Greece)·2026
Same author

Preoperative Botulinum Toxin A Within an Endoscopic Retromuscular Strategy for Ventral Hernia Repair: A Preliminary Prospective Study in Japanese Patients.

Asian journal of endoscopic surgery·2026
Same author

Association Between Chin-Tuck-Generated Force and Sarcopenia in Community-Dwelling Older Adults: A Cross-Sectional Study.

Healthcare (Basel, Switzerland)·2026
Same author

Association Between Floor of Residence and Frailty in Walk-Up Buildings Among Functionally Independent Older Adults: A Cross-Sectional Study.

Healthcare (Basel, Switzerland)·2026
Same author

Comparison of synchronous and nonsynchronous imitation in robotic surgical training: a pilot randomized-controlled trial.

Surgery today·2025
Same author

Long-Term Survival of a Patient With Locally Advanced Esophageal Cancer Treated With Chemoradiotherapy Following Esophageal Bypass Surgery: A Case Report.

Cureus·2025
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:

  • Computational Neuroscience
  • Artificial Neural Networks
  • Visual Processing

Background:

  • The middle temporal (MT) and medial superior temporal (MST) areas of the brain are crucial for processing visual motion.
  • Understanding the neural mechanisms underlying motion perception is a key challenge in neuroscience.
  • Artificial neural networks offer a powerful framework for modeling biological neural systems.

Purpose of the Study:

  • To develop and evaluate two-layered neural network models simulating MT and MST neurons.
  • To investigate the impact of neuron state complexity (Q-states) on the models' ability to represent responses to optical flow.
  • To compare the performance of binary (Q=2) and multi-state (Q>=3) neuron models.

Main Methods:

  • Implementation of two-layered neural network models incorporating Q-state neurons (Q>=2).

Related Experiment Videos

  • Utilizing the wake-sleep algorithm, a local learning rule, for model training.
  • Simulation of MST neuron responses to various optical flow patterns using both binary and multi-state neuron models.

    • Main Results:

    • The binary neuron model (Q=2) was investigated for its response properties to optical flow.
    • Extension to Q-state neurons (Q>=3) demonstrated enhanced response properties compared to the binary model.
    • Increasing the number of states (Q) in neurons led to improved representation of MST neuron responses.

    Conclusions:

    • Two-layered neural network models with Q-state neurons can effectively simulate visual motion processing.
    • Multi-state neurons (Q>=3) provide a more accurate and robust representation of MST neuron responses to optical flow than binary neurons.
    • The wake-sleep algorithm's local learning rules are suitable for training these complex neural network models.