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 Video

Updated: Jun 27, 2025

Author Spotlight: Enhancement of Salient Object Detection for Smart Grid Applications
03:31

Author Spotlight: Enhancement of Salient Object Detection for Smart Grid Applications

Published on: December 15, 2023

524

Latency-Aware Unified Dynamic Networks for Efficient Image Recognition.

Yizeng Han, Zeyu Liu, Zhihang Yuan

    IEEE Transactions on Pattern Analysis and Machine Intelligence
    |April 25, 2024
    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

    Integrated Single-Cell and Bulk RNA Sequencing Identifies Macrophage Heterogeneity and Mitophagy-Related Biomarkers in Idiopathic Pulmonary Fibrosis.

    International journal of molecular sciences·2026
    Same author

    Study on Synthesis and Performance of a Hybrid Crosslinked Composite Gel for High-Temperature Lost Circulation Control.

    Gels (Basel, Switzerland)·2026
    Same author

    Neuropsychiatric systemic lupus erythematosus subtypes identified by unsupervised clustering: A single-center cohort study.

    Rheumatology and immunology research·2026
    Same author

    Design of a miniature sensing module for pressure mapping in functionality hydrogel using programmable system-on-chip.

    HardwareX·2026
    Same author

    Effect of a checklist-based core competency training and evaluation system on work-related alienation among critical care nurses.

    Frontiers in public health·2026
    Same author

    An annotated genome of freshwater amphipod (Gammarus nekkensis).

    Scientific data·2026
    Same journal

    A Unified and Fast-Sampling Diffusion Bridge Framework via Stochastic Optimal Control.

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

    Robust 3D Semantic Occupancy Prediction With Calibration-Free Spatial Transformation.

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

    Image Restoration via Multi-domain Learning.

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

    A Comprehensive Survey on Multimodal Recommender Systems: Taxonomy, Evaluation, and Future Directions.

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

    Benchmarking the Robustness of Autonomous Driving to Environmental Illusions: A Lane Perception Perspective.

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

    Learning Topology-Aware Representations via Test-Time Adaptation for Anomaly Segmentation.

    IEEE transactions on pattern analysis and machine intelligence·2026
    See all related articles

    Latency-Aware Unified Dynamic Networks (LAUDNet) improve deep learning efficiency by unifying dynamic inference paradigms and optimizing hardware scheduling. This framework significantly reduces practical latency while maintaining accuracy across various vision tasks.

    Area of Science:

    • Deep Learning
    • Computer Vision
    • Artificial Intelligence

    Background:

    • Dynamic networks selectively activate computation units or allocate resources to information-rich regions, reducing unnecessary computations.
    • Practical efficiency of dynamic models often lags behind theoretical potential due to fragmented frameworks, inadequate scheduling strategies, and complex latency evaluation.

    Purpose of the Study:

    • To introduce Latency-Aware Unified Dynamic Networks (LAUDNet), a general framework addressing the practical efficiency gap in dynamic deep learning models.
    • To integrate spatially-adaptive computation, layer skipping, and channel skipping into a unified formulation.
    • To enhance scheduling optimization through a hardware-aware latency predictor.

    Main Methods:

    • Developed LAUDNet, a unified framework integrating three dynamic inference paradigms: spatially-adaptive computation, layer skipping, and channel skipping.

    More Related Videos

    Deep Neural Networks for Image-Based Dietary Assessment
    13:19

    Deep Neural Networks for Image-Based Dietary Assessment

    Published on: March 13, 2021

    9.1K
    A Swin Transformer-Based Model for Thyroid Nodule Detection in Ultrasound Images
    04:23

    A Swin Transformer-Based Model for Thyroid Nodule Detection in Ultrasound Images

    Published on: April 21, 2023

    1.8K

    Related Experiment Videos

    Last Updated: Jun 27, 2025

    Author Spotlight: Enhancement of Salient Object Detection for Smart Grid Applications
    03:31

    Author Spotlight: Enhancement of Salient Object Detection for Smart Grid Applications

    Published on: December 15, 2023

    524
    Deep Neural Networks for Image-Based Dietary Assessment
    13:19

    Deep Neural Networks for Image-Based Dietary Assessment

    Published on: March 13, 2021

    9.1K
    A Swin Transformer-Based Model for Thyroid Nodule Detection in Ultrasound Images
    04:23

    A Swin Transformer-Based Model for Thyroid Nodule Detection in Ultrasound Images

    Published on: April 21, 2023

    1.8K
  • Incorporated a latency predictor to optimize scheduling strategies and accurately estimate inference latency on specific hardware.
  • Evaluated LAUDNet across image classification, object detection, and instance segmentation tasks.

    • Main Results:

    • LAUDNet significantly bridges the gap between theoretical and practical efficiency for dynamic networks.
    • Achieved over 50% practical latency reduction for ResNet-101 on hardware platforms like V100, RTX 3090, and TX2 GPUs.
    • Demonstrated a superior accuracy-efficiency trade-off compared to existing methods.

    Conclusions:

    • LAUDNet provides a unified and efficient solution for dynamic network inference, improving hardware utilization.
    • The framework effectively reduces latency and enhances performance across diverse computer vision applications.
    • LAUDNet represents a significant advancement in optimizing the practical deployment of dynamic deep learning models.