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

CKD: Contrastive Knowledge Distillation for Cross-Dataset EEG Classification.

Ziwei Wang, Xingyi He, Hongbin Wang

    IEEE Transactions on Bio-Medical Engineering
    |June 8, 2026
    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

    fastSeizureNet: Accurate and efficient knowledge-data fusion for semi-supervised seizure detection.

    Neural networks : the official journal of the International Neural Network Society·2026
    Same author

    SACM: SEEG-Audio Contrastive Matching for Chinese Speech Decoding.

    IEEE transactions on bio-medical engineering·2026
    Same author

    Mirror Descent Safe Policy Optimization for Reinforcement Learning Agents.

    IEEE transactions on pattern analysis and machine intelligence·2026
    Same author

    Electroencephalography Enables Continuous Decoding of Hand Motion Angles in Polar Coordinates.

    Cyborg and bionic systems (Washington, D.C.)·2026
    Same author

    Discovery of the reward function for embodied reinforcement learning agents.

    Nature communications·2025
    Same author

    Magnetoencephalography (MEG) based non-invasive Chinese speech decoding.

    Journal of neural engineering·2025
    Same journal

    Magnetic Resonance Spectroscopy Deep Learning with Magnetic Resonance Background Generator Enables In Vivo Metabolite Quantification of Hepatic Encephalopathy.

    IEEE transactions on bio-medical engineering·2026
    Same journal

    Use of RPNIs and Implanted Electrodes for Prosthetic Wrist and Multi-Grip Hand Control during Functional Tasks: A Case Study.

    IEEE transactions on bio-medical engineering·2026
    Same journal

    Healthy Limb Driven Prediction for Real Time Control of Unilateral Exoskeletons in Gait Rehabilitation.

    IEEE transactions on bio-medical engineering·2026
    Same journal

    A Miniature Wearable Ultrasound System for Continuous Bladder Monitoring with Sleeping-Position-Robust Modeling Strategies.

    IEEE transactions on bio-medical engineering·2026
    Same journal

    A Bi-objective Array Optimization Framework for Magnetocardiographic Source Imaging.

    IEEE transactions on bio-medical engineering·2026
    Same journal

    A Dynamic Mutual Information Measure of Phase-Amplitude Coupling with Uncertainty Quantification.

    IEEE transactions on bio-medical engineering·2026
    See all related articles

    This study introduces a contrastive knowledge distillation (CKD) framework to enhance cross-dataset electroencephalography (EEG) decoding. CKD improves brain-computer interface (BCI) model generalizability by aligning feature representations across diverse EEG datasets.

    Area of Science:

    • Neuroscience
    • Machine Learning
    • Biomedical Engineering

    Background:

    • Cross-dataset transfer in electroencephalography (EEG) brain-computer interfaces (BCIs) is hindered by significant data distribution shifts.
    • Variations in subjects, devices, and protocols create challenges for reliable EEG decoding across different datasets.

    Purpose of the Study:

    • To develop an advanced framework for improving cross-dataset EEG decoding by enhancing knowledge transfer.
    • To address the limitations of conventional output-level distillation in handling distribution shifts.

    Main Methods:

    • A two-stage contrastive knowledge distillation (CKD) framework was proposed, involving teacher pretraining and online adaptation.
    • CKD integrates logit-level distillation with feature-level contrastive alignment to transfer both predictive behavior and representation structure.

    Related Experiment Videos

  • The framework aims to improve knowledge transfer beyond traditional methods.

    • Main Results:

    • CKD significantly outperformed twelve baseline methods across five motor imagery EEG datasets in both single-source and multi-source transfer scenarios.
    • The framework demonstrated superior teacher-student alignment in feature geometry and distribution consistency.
    • Explicit domain adaptation further enhanced CKD's performance.

    Conclusions:

    • The proposed CKD framework offers an effective solution for robust cross-dataset EEG decoding.
    • CKD enhances predictive knowledge transfer and latent feature alignment, crucial for overcoming severe dataset shifts.
    • This work advances the robustness and generalizability of EEG decoding for practical BCI applications in real-world conditions.