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 Video

Updated: Oct 18, 2025

Author Spotlight: Enhancement of Salient Object Detection for Smart Grid Applications
03:31

Author Spotlight: Enhancement of Salient Object Detection for Smart Grid Applications

Published on: December 15, 2023

692

An adiabatic method to train binarized artificial neural networks.

Yuansheng Zhao1,2, Jiang Xiao3,4,5

  • 1Department of Physics and State Key Laboratory of Surface Physics, Fudan University, Shanghai, 200433, China.

Scientific Reports
|October 6, 2021
PubMed
Summary

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

Synchrotron x-ray diffraction study of liquid and glassy toluene.

The Journal of chemical physics·2026
Same author

Population Pharmacokinetics of Oral Gecacitinib in Healthy Subjects and Patients with Autoimmune and Inflammatory Diseases.

Journal of clinical pharmacology·2026
Same author

Durable but impaired humoral and cellular immune responses against monkeypox virus in people with HIV over 18 months.

Journal of translational medicine·2026
Same author

Characterization of telomere-related gene subtypes in lung adenocarcinoma and their implications for prognosis and treatment.

Discover oncology·2026
Same author

Metabolic outcomes of bictegravir/emtricitabine/tenofovir alafenamide versus dolutegravir/lamivudine in treatment-naïve people living with HIV: a 48-week retrospective study.

Expert review of anti-infective therapy·2026
Same author

Association between systemic inflammation-immune index and comorbidities in patients with ankylosing spondylitis: a cross-sectional study.

Clinical rheumatology·2026

This study introduces an adiabatic training method to binarize artificial neural networks, replacing complex operations with simpler ones. This approach maintains performance while significantly boosting computational efficiency for easier deployment.

Area of Science:

  • Computer Science
  • Artificial Intelligence
  • Machine Learning

Background:

  • Artificial neural networks (ANNs) rely on neurons and synapses with non-linear activation functions and real-valued weights.
  • Multiplication and Accumulate (MAC) operations are computationally intensive in ANNs.
  • Binarizing neuron outputs and/or weights can replace MAC operations with efficient XNOR operations.

Purpose of the Study:

  • To demonstrate an adiabatic training method for binarizing fully-connected and convolutional neural networks.
  • To evaluate the performance of binarized neural networks against conventional networks.
  • To assess the applicability of the adiabatic method across diverse network architectures and tasks.

Main Methods:

  • Developed an adiabatic training methodology to convert neural network weights and activations to binary values.

More Related Videos

Decoding Natural Behavior from Neuroethological Embedding
08:00

Decoding Natural Behavior from Neuroethological Embedding

Published on: October 3, 2025

87
Artificial Intelligence Approaches to Assessing Primary Cilia
08:58

Artificial Intelligence Approaches to Assessing Primary Cilia

Published on: May 1, 2021

3.8K

Related Experiment Videos

Last Updated: Oct 18, 2025

Author Spotlight: Enhancement of Salient Object Detection for Smart Grid Applications
03:31

Author Spotlight: Enhancement of Salient Object Detection for Smart Grid Applications

Published on: December 15, 2023

692
Decoding Natural Behavior from Neuroethological Embedding
08:00

Decoding Natural Behavior from Neuroethological Embedding

Published on: October 3, 2025

87
Artificial Intelligence Approaches to Assessing Primary Cilia
08:58

Artificial Intelligence Approaches to Assessing Primary Cilia

Published on: May 1, 2021

3.8K
  • Applied the method to fully-connected and convolutional neural network architectures.
  • Tested the binarized networks on tasks including handwriting recognition, cat-dog recognition, audio recognition, and image classification (CIFAR-10).

    • Main Results:

    • Binarized neural networks achieved performance nearly identical to conventional networks with real-valued weights and activations.
    • The adiabatic training method requires minimal modifications to existing training algorithms.
    • The method was successfully applied to various network types like ResNet-20 and VGG-Small.

    Conclusions:

    • The adiabatic training method effectively binarizes neural networks without altering structure or size.
    • This approach significantly enhances computational efficiency and simplifies neural network deployment.
    • The method is versatile and applicable to diverse neural network architectures and tasks.