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

Character recognition using a dynamic optoelectronic neural network with unipolar binary weights.

M Oita, M Takahashi, S Tai

    Optics Letters
    |September 23, 2009
    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

    SU-E-T-106: Evaluation of Sensitivity and Uniformity of New Radiochromic Film with Two Commercial Scanners.

    Medical physics·2017
    Same author

    Optical learning neurochip with internal analog memory.

    Applied optics·2010
    Same author

    Optoelectronic associative memory using an advanced optical neurochip.

    Applied optics·2010
    Same author

    Optical implementation of an associative neural network model with a stochastic process.

    Applied optics·2010
    Same author

    Laser Doppler velocimeter with a novel optical fiber probe.

    Applied optics·2010
    Same author

    N x N to N pattern associative memory by using optoelectronic neurochips.

    Optics letters·2009
    Same journal

    Gaussian-modulated continuous-variable quantum key distribution over 60 km fiber using an integrated silicon photonic receiver.

    Optics letters·2026
    Same journal

    E2E-OCT: end-to-end joint learning model using optical coherence tomography images for vocal cord leukoplakia diagnosis.

    Optics letters·2026
    Same journal

    Holographic generation of panoramic 3D scenes by concave ellipsoidal mirror reflection.

    Optics letters·2026
    Same journal

    Dual-pilot phase recovery with pair-wise maximum-ratio combining for coherent PONs.

    Optics letters·2026
    Same journal

    Mapping the whispering gallery modes of a CaF<sub>2</sub> disk resonator with half-tapered fibers to estimate the fundamental mode volume.

    Optics letters·2026
    Same journal

    Quantitative estimation of deep-subwavelength scale via dark-field scattering axial energy concentration decay profiles.

    Optics letters·2026
    See all related articles

    Researchers developed a new quantized learning rule for optical neural networks using binary weights. This method improves character recognition in optoelectronic networks by addressing spatial light modulator limitations.

    Area of Science:

    • Optoelectronics
    • Artificial Intelligence
    • Computer Science

    Background:

    • Optical neural networks require efficient learning rules for implementation.
    • Spatial light modulators (SLMs) used as synaptic devices suffer from insufficient contrast ratios, impacting performance.
    • Existing learning rules may not be optimal for unipolar binary weight systems in optical settings.

    Purpose of the Study:

    • To introduce a novel quantized learning rule suitable for optical neural network implementation.
    • To propose an input-dependent thresholding method to mitigate SLM contrast ratio issues.
    • To experimentally validate the proposed methods for character recognition.

    Main Methods:

    • Development of a quantized learning rule utilizing unipolar binary weights.

    Related Experiment Videos

  • Implementation of an input-dependent thresholding operation.
  • Experimental setup using a single set of an optoelectronic three-layered network for character recognition.

    • Main Results:

    • The proposed quantized learning rule enables effective training of optical neural networks.
    • The input-dependent thresholding successfully compensates for insufficient contrast ratios in SLMs.
    • Experimental demonstration of accurate recognition of 26 alphabet characters using the developed system.

    Conclusions:

    • The novel quantized learning rule and thresholding method are effective for optical neural networks.
    • This approach offers a viable solution for practical optical implementation of neural networks.
    • The experimental results confirm the potential of this technology for real-world applications like character recognition.