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

Varieties of Helmholtz Machine.

Geoffrey E. Hinton1, Peter Dayan

  • 1University of Toronto, USA

Neural Networks : the Official Journal of the International Neural Network Society
|November 1, 1996
PubMed
Summary
This summary is machine-generated.

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

Latent subdimensions of anxiety and depression differentially influence exertion of effort in pursuit of reward versus avoidance of threat.

Translational psychiatry·2026
Same author

Neural signatures of model-based and model-free reinforcement learning across prefrontal cortex and striatum.

eLife·2026
Same author

Uncertainty for better and worse.

Current opinion in neurobiology·2026
Same author

Dopamine dynamics in human anterior cingulate cortex during Pavlovian-instrumental conflict.

bioRxiv : the preprint server for biology·2026
Same author

Modality-general sensitivity of pupil responses to regularity violations.

Cognitive, affective & behavioral neuroscience·2026
Same author

Metacognitive efficiency in learned value-based choice.

PLoS computational biology·2026

Helmholtz machines are novel unsupervised learning architectures. This paper explores variations and their links to brain information processing using a wake-sleep algorithm.

Area of Science:

  • Artificial Intelligence
  • Computational Neuroscience
  • Machine Learning

Background:

  • Unsupervised learning architectures are crucial for modeling complex data distributions.
  • Hierarchical probabilistic models offer a framework for understanding sensory processing.
  • The need for efficient learning algorithms in neural networks is paramount.

Purpose of the Study:

  • To introduce and explore variations of the Helmholtz machine, a novel unsupervised learning architecture.
  • To detail the wake-sleep learning algorithm used by Helmholtz machines.
  • To investigate the relationship between Helmholtz machines and cortical information processing.

Main Methods:

  • Utilized top-down connections for building probability density models.

Related Experiment Videos

  • Employed bottom-up connections for constructing inverse models.
  • Applied the purely local delta rule within the wake-sleep learning algorithm.

    • Main Results:

    • Demonstrated the feasibility of Helmholtz machines for unsupervised learning.
    • Introduced several distinct varieties of Helmholtz machines.
    • Established potential connections between these machine architectures and biological neural systems.

    Conclusions:

    • Helmholtz machines represent a promising new direction in unsupervised learning.
    • The wake-sleep algorithm offers an efficient, local learning mechanism.
    • Further research into Helmholtz machines could illuminate principles of cortical computation.