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

On Temporal Generalization of Simple Recurrent Networks.

Stanley C. Ahalt1, Liu Xiaomei, Wang DeLiang

  • 1The Ohio State University, Ohio, USA

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

Simple recurrent networks exhibit limitations in temporal generalization, showing interval invariance but not rate invariance for complex tasks. These findings suggest significant constraints in their ability to process time-varying information, impacting applications like speech processing.

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

Transcriptome profile and its partly verification of human hepatocellular carcinoma cells exposed to Yuzhizi () seed extract.

Journal of traditional Chinese medicine = Chung i tsa chih ying wen pan·2022
Same author

Lung hypoplasia in rats with esophageal atresia and tracheo-esophageal fistula.

Pediatric research·2012
Same author

Microscopic magnetic resonance imaging of the thoracic venous system in rats with congenital diaphragmatic hernia.

Pediatric surgery international·2010
Same author

Abnormal development of the enteric nervous system in rat embryos and fetuses with congenital diaphragmatic hernia.

Pediatric surgery international·2010
Same author

Phytoextraction of Zn and Cu from sewage sludge and impact on agronomic characteristics.

Journal of environmental science and health. Part A, Toxic/hazardous substances & environmental engineering·2005

Area of Science:

  • Artificial Intelligence
  • Computational Neuroscience
  • Machine Learning

Background:

  • Simple recurrent networks, specifically Elman networks, are commonly employed for temporal processing tasks.
  • Understanding the temporal generalization capabilities of these networks is crucial for advancing AI applications.
  • Comparing network performance to human temporal processing offers insights into biological and artificial systems.

Purpose of the Study:

  • To investigate the temporal generalization abilities of Elman networks.
  • To compare the temporal processing capabilities of Elman networks with human performance.
  • To identify limitations in Elman networks regarding rate invariance and temporal generalization.

Main Methods:

  • Elman networks were trained to generate temporal trajectories at various sampling rates.

Related Experiment Videos

  • Networks were tested using trajectories at trained and novel sampling rates, including mixed rates.
  • Performance was evaluated on both synthetic trajectories and measured speech data.

    • Main Results:

    • For simple trajectories, Elman networks demonstrated interval invariance but lacked rate invariance.
    • Complex trajectories requiring contextual information showed no significant temporal generalization in the networks.
    • Results with speech data mirrored findings from synthetic trajectories, indicating consistent limitations.

    Conclusions:

    • Elman networks exhibit severe limitations in temporal generalization, particularly concerning rate invariance.
    • The findings suggest that current simple recurrent network architectures are insufficient for robust temporal processing in complex scenarios.
    • Further research is needed to develop neural network architectures capable of achieving true rate invariance and enhanced temporal generalization.