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

Symmetric Image-Text Tuning With Entropy-Guided Fusion for Online Continual Learning in Non-Stationary Visual

Leyuan Wang, Liuyu Xiang, Yujie Wei

    IEEE Transactions on Image Processing : a Publication of the IEEE Signal Processing Society
    |March 23, 2026
    PubMed
    Summary

    Related Concept Videos

    Convolution: Math, Graphics, and Discrete Signals01:24

    Convolution: Math, Graphics, and Discrete Signals

    1.1K
    In any LTI (Linear Time-Invariant) system, the convolution of two signals is denoted using a convolution operator, assuming all initial conditions are zero. The convolution integral can be divided into two parts: the zero-input or natural response and the zero-state or forced response, with t0 indicating the initial time.
    To simplify the convolution integral, it is assumed that both the input signal and impulse response are zero for negative time values. The graphical convolution process...
    1.1K

    You might also read

    Related Articles

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

    Sort by
    Same author

    Vancomycin-stressed <i>Staphylococcus aureus</i> extracellular vesicles hijack macrophage autophagy to enhance bacterial intracellular survival.

    Autophagy·2026
    Same author

    Predictors of rapidly progressive interstitial lung disease in anti-MDA5 dermatomyositis: a meta-analysis.

    Rheumatology (Oxford, England)·2026
    Same author

    Mechanism of <i>Inonotus hispidus</i> in suppressing renal cell carcinoma proliferation via regulation of the PI3K/AKT/mTOR pathway.

    Frontiers in pharmacology·2026
    Same author

    Outcomes of an optimized ciclosporin-free haploidentical HSCT protocol in paediatric patients with cerebral adrenoleukodystrophy.

    British journal of haematology·2026
    Same author

    Cathepsin C orchestrates RSV-induced asthma exacerbation through the dual effect of monocyte-derived macrophages.

    Cellular & molecular immunology·2026
    Same author

    Enhanced removal of estradiol by Twice-ball milled magnetic biochar: Mechanisms and efficiency.

    Journal of environmental management·2026

    This study addresses catastrophic forgetting in online continual learning for CLIP models by introducing a symmetric image-text tuning strategy. This method balances adapting to new data with preserving prior knowledge in realistic scenarios.

    Area of Science:

    • Computer Science
    • Artificial Intelligence
    • Machine Learning

    Background:

    • Online continual learning (OCL) models learn from non-stationary data streams.
    • CLIP models show asymmetric image-text interactions in OCL, causing catastrophic forgetting.
    • Previous methods struggle with preserving knowledge while adapting to new data.

    Purpose of the Study:

    • To mitigate catastrophic forgetting in CLIP models during OCL.
    • To develop a strategy for balancing knowledge adaptation and preservation.
    • To introduce a realistic OCL benchmark for evaluating model performance.

    Main Methods:

    • Proposed Symmetric Image-Text Tuning (SIT) to remove asymmetric text supervision.
    • Introduced Entropy-Guided Fusion (EGF) for adaptive prediction combination.

    Related Experiment Videos

  • Developed the MiD-Blurry benchmark with diverse class distributions and blurred boundaries.

    • Main Results:

    • SIT and EGF effectively reduce catastrophic forgetting in CLIP models.
    • The approach maintains a balance between learning new information and retaining old knowledge.
    • Experiments demonstrate strong performance on standard benchmarks and the new MiD-Blurry setting.

    Conclusions:

    • The proposed SIT and EGF strategy offers an effective solution for OCL with CLIP models.
    • The MiD-Blurry benchmark provides a more realistic evaluation for OCL systems.
    • This work advances OCL by improving model plasticity and stability.