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

A Hierarchical Mixture-Of-Experts Framework for Few Labeled Node Classification.

Yimeng Wang1, Zhiyao Yang1, Xiangjiu Che1

  • 1College of Computer Science and Technology, Jilin University, Changchun 130012, Jilin, China; Key Laboratory of Symbolic Computation and Knowledge Engineering of MOE, Jilin University, Changchun 130012, Jilin, China.

Neural Networks : the Official Journal of the International Neural Network Society
|March 29, 2025
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

Association of Pre-Transplant Tyrosine Kinase Inhibitor Therapy With Tumor Response and Survival in Hepatocellular Carcinoma: A 10-Year Retrospective Cohort Study of 427 Patients.

World journal of surgery·2026
Same author

Multimodal graph neural network with large language models for node and link prediction.

Frontiers in artificial intelligence·2026
Same author

HO-1 alleviates lipopolysaccharide-induced acute lung injury in mice by downregulating TFE3 expression and nuclear translocation and suppressing Golgi stress response.

International immunopharmacology·2026
Same author

Perioperative tislelizumab for early-stage hepatocellular carcinoma: A phase II trial with integrated tumour microenvironment profiling and predictive modeling.

Drug resistance updates : reviews and commentaries in antimicrobial and anticancer chemotherapy·2026
Same author

Molecular-Anion Interfacial Engineering for Solid Electrolyte Interphase to Obtain High-Performance Aqueous Zinc-Ion Batteries.

Chemistry (Weinheim an der Bergstrasse, Germany)·2026
Same author

Exploring the Role of Skin Microbiota in Autoimmune Skin Diseases from a Bidirectional Mendelian Randomization Perspective.

Clinical, cosmetic and investigational dermatology·2026

This study introduces Hierarchical Mixture-of-Experts (HMoE) to improve Few Labeled Node Classification (FLNC) by reducing overfitting and enhancing feature representation. HMoE achieves better performance on graph data with limited labels.

Area of Science:

  • Graph Neural Networks
  • Machine Learning
  • Data Science

Background:

  • Few Labeled Node Classification (FLNC) is challenging due to extremely limited training nodes.
  • Graph Neural Networks (GNNs) struggle with feature convergence in FLNC.
  • Mixture of Experts (MoE) can overfit when applied directly to FLNC.

Purpose of the Study:

  • To propose a novel framework, Hierarchical Mixture-of-Experts (HMoE), to address overfitting and feature convergence in FLNC.
  • To enhance feature representation for graph data with limited labels.
  • To improve the performance of node classification tasks under data scarcity.

Main Methods:

  • Implemented three data augmentation techniques to enrich input features and mitigate overfitting.
  • Designed a hierarchical mixture-of-experts encoder with distinct layers for unique and shared feature extraction.
Keywords:
Data augmentationFew labeled graphMixture of expertsNode classification

Related Experiment Videos

  • Incorporated an auxiliary task with a gradient reversal mechanism to improve feature representation ability.

    • Main Results:

    • The proposed HMoE framework demonstrated superior performance compared to baseline methods.
    • Achieved an average performance improvement of 1.2% across six diverse datasets.
    • Effectively reduced overfitting and improved feature representation for FLNC.

    Conclusions:

    • HMoE framework offers a robust solution for Few Labeled Node Classification.
    • The combination of hierarchical experts, data augmentation, and auxiliary tasks enhances GNN performance in low-label scenarios.
    • This approach advances the capabilities of GNNs for real-world graph data analysis with limited supervision.