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

FinePruner: Unbiased Attention-Head-Level Fine-Grained Token Reduction for Efficient Inference of Large

Zishuo Wang, Xiangtian Zheng, Yuxin Peng

    IEEE Transactions on Image Processing : a Publication of the IEEE Signal Processing Society
    |May 5, 2026
    PubMed
    Summary
    This summary is machine-generated.

    Related Concept Videos

    Improving Translational Accuracy02:07

    Improving Translational Accuracy

    2.7K
    2.7K
    Improving Translational Accuracy02:07

    Improving Translational Accuracy

    11.6K
    Base complementarity between the three base pairs of mRNA codon and the tRNA anticodon is not a failsafe mechanism. Inaccuracies can range from a single mismatch to no correct base pairing at all. The free energy difference between the correct and nearly correct base pairs can be as small as 3 kcal/ mol. With complementarity being the only proofreading step, the estimated error frequency would be one wrong amino acid in every 100 amino acids incorporated. However, error frequencies observed in...
    11.6K

    You might also read

    Related Articles

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

    Sort by
    Same author

    Effects of serotonergic pharmacological manipulation on exercise performance and central fatigue: a systematic review and three-level meta-analysis.

    Journal of the International Society of Sports Nutrition·2026
    Same author

    The cellular and molecular mechanisms of linalool in cancer therapy: Recent advances.

    Fitoterapia·2026
    Same author

    Comparative chloroplast genomics of Verbenaceae: structural divergence, adaptive evolution, and phylogenomic insights.

    Planta·2026
    Same author

    [Exploration and practice of cultivating top-notch innovative talents in biological sciences under the dual-track framework integrating "Strengthening Basic Disciplines Program" and "Top-notch Student Training Program"].

    Sheng wu gong cheng xue bao = Chinese journal of biotechnology·2026
    Same author

    Global trends in the burden of inflammatory bowel disease from 1990 to 2021: socioeconomic and sex-specific disparities with exploratory mendelian randomization analyses.

    Clinical and experimental medicine·2026
    Same author

    Improving plant salt tolerance using the endophytic halophile Vreelandella salis sp. nov.

    Plant physiology and biochemistry : PPB·2026

    Large Vision-Language Models (LVLMs) face computational challenges due to numerous visual tokens. FinePruner reduces these tokens by exploring attention patterns, improving efficiency without sacrificing performance in fine-grained tasks.

    Area of Science:

    • Computer Vision
    • Artificial Intelligence
    • Machine Learning

    Background:

    • Large Vision-Language Models (LVLMs) incur high computational costs from attention mechanisms with many visual tokens.
    • Existing token reduction methods struggle with attention biases, leading to performance degradation or limited acceleration, especially in fine-grained perception tasks.

    Purpose of the Study:

    • To propose an unbiased fine-grained token reduction method, FinePruner, to mitigate attention biases in LVLMs.
    • To enhance the accuracy-efficiency tradeoff for LVLMs in fine-grained visual tasks.

    Main Methods:

    • FinePruner explores attention patterns at the attention-head level to overcome attention biases.
    • It employs a two-stage pipeline: Instruction-Agnostic Clustering to group tokens and Attention-Refined Pruning to identify critical tokens based on less biased attention heads.

    Related Experiment Videos

    Main Results:

    • Comparative studies confirmed the importance of preserving object-related tokens.
    • Visualizations revealed layer- and head-specific attention bias patterns in LVLMs.
    • FinePruner demonstrated superior accuracy-efficiency tradeoffs compared to state-of-the-art methods on VQA and fine-grained benchmarks.

    Conclusions:

    • FinePruner effectively reduces visual tokens in LVLMs by addressing attention biases.
    • The method offers a significant improvement in balancing accuracy and computational efficiency for fine-grained perception tasks.