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

Robust zero-shot learning with ambiguous labels via visual-semantic alignment and dynamic disambiguation.

Jiangnan Li1, Xiaowen Yan1, Linqing Huang2

  • 1College of Computer Science and Technology, Qingdao University, Qingdao, 266071, China.

Neural Networks : the Official Journal of the International Neural Network Society
|May 23, 2026
PubMed
Summary

Related Concept Videos

Associative Learning01:27

Associative Learning

Associative learning is a fundamental concept in behavioral psychology, wherein a connection is established between two stimuli or events, leading to a learned response. This process is critical in understanding how behaviors are acquired and modified. Conditioning, the mechanism through which associations are formed, can be divided into two main types: classical conditioning and operant conditioning, each elucidating different aspects of associative learning.
Classical conditioning, also known...

You might also read

Related Articles

Articles linked to this work by shared authors, journal, and citation graph.

Sort by
Same author

Short-Chain Fatty Acid Metabolism in Acute Lung Injury and Intervention Strategies from Traditional Chinese Medicine.

Journal of ethnopharmacology·2026
Same author

Dual Joint Covariance Alignment Method for Incomplete Data Classification.

IEEE transactions on neural networks and learning systems·2026
Same author

Genome-Wide Identification and Expression Profiling of the <i>ARF</i> Gene Family During Seed Germination in Sesame (<i>Sesamum indicum</i> L.) Under Abiotic Stresses.

International journal of molecular sciences·2026
Same author

Comparative mitochondrial genomics of cultivated <i>Sesamum indicum</i> and its wild relative <i>Sesamum schinzianum</i> provides insights into structural features and organellar evolution.

Frontiers in plant science·2026
Same author

Noninvasive Analysis of Skin Emanations during Cupping Therapy by Thin-Film Solid-Phase Microextraction and Dielectric Barrier Discharge Ionization Mass Spectrometry.

Analytical chemistry·2026
Same author

Voltage-Triggered Emergent Dynamics in Strongly Coupled Nanomagnet Networks for Neuromorphic Computing.

ACS nano·2026

This study introduces a robust zero-shot learning (ZSL) framework to handle ambiguous labels. The Dynamic Visual-semantic Alignment (DVSA) method improves recognition accuracy by aligning visual and semantic data despite noisy labels.

Area of Science:

  • Computer Science
  • Artificial Intelligence
  • Machine Learning

Background:

  • Zero-shot Learning (ZSL) excels at recognizing new classes without prior examples.
  • Real-world ZSL is hindered by label noise and ambiguity, degrading performance.
  • Existing ZSL methods often assume clean labels, failing in practical scenarios.

Purpose of the Study:

  • To develop a robust zero-shot learning framework capable of handling ambiguous labels.
  • To enhance the performance of ZSL models in the presence of noisy or uncertain training data.
  • To improve the generalization ability of ZSL models to unseen classes under ambiguous supervision.

Main Methods:

  • Proposed a unified framework, Dynamic Visual-semantic Alignment (DVSA).
  • Integrated bidirectional visual-semantic alignment, attribute-level Mutual Information (MI) regularization, and dynamic label disambiguation.
Keywords:
Ambiguous labelsDynamic disambiguationMutual information estimationZero-shot learning

Related Experiment Videos

  • Employed attention mechanisms for visual-semantic calibration and contrastive optimization for separable embeddings.

    • Main Results:

    • DVSA demonstrated improved cross-modal correspondence under ambiguous supervision.
    • Attribute-level MI regularization enhanced the separability of embeddings.
    • Dynamic label disambiguation progressively refined supervision signals, reducing the instance-label semantic gap.
    • The framework showed effectiveness and robustness across multiple benchmark datasets.

    Conclusions:

    • The proposed DVSA framework effectively addresses the challenge of ambiguous labels in zero-shot learning.
    • DVSA enables more reliable transfer learning to unseen classes even with noisy training data.
    • The integrated components contribute to a more robust and accurate ZSL system in practical applications.