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

Understanding the Performance of Deep Computer Vision Models: A Symbolic Regression Approach to Accuracy and Latency

Divyesh Rameshbhai Dhanani1, Faraz Kayani1, Saif U Din1

  • 1Computer Vision Lab, Center for Artificial Intelligence and Data Science (CAIDAS) and Institute of Computer Science (IFI), University of Würzburg, John-Skilton-Straße 4a, 97074 Würzburg, Germany.

Sensors (Basel, Switzerland)
|July 15, 2026
PubMed
Summary

Related Concept Videos

Improving Translational Accuracy02:07

Improving Translational Accuracy

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...
Improving Translational Accuracy02:07

Improving Translational Accuracy

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...

You might also read

Related Articles

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

Sort by
Same author

Applying artificial intelligence to assess the impact of orthognathic treatment on gender-affirming facial features.

European journal of orthodontics·2026
Same author

Efficient End-to-End Convolutional Architecture for Point-of-Gaze Estimation.

Journal of imaging·2024
See all related articles

This study introduces a statistical framework to predict deep vision model accuracy and inference latency on edge hardware. Batch size and layer count are key accuracy predictors, while learning rate shows a non-monotonic relation, enabling better model screening.

Area of Science:

  • Computer Vision
  • Machine Learning
  • Statistical Modeling

Background:

  • Deploying deep vision models on edge hardware is challenging due to poorly understood relationships between architecture, hyperparameters, accuracy, and inference latency.
  • Existing methods lack systematic, data-driven characterization of these crucial factors.

Purpose of the Study:

  • To develop a statistical framework for interpretable, closed-form insights into deep vision model accuracy and inference latency.
  • To identify key architectural and training hyperparameters influencing model performance on edge devices.

Main Methods:

  • Utilized distance correlation (dCor) and maximal information coefficient (MIC) across seven image-classification datasets.
  • Applied PySR symbolic regression for the first time to cross-dataset vision model accuracy prediction, deriving interpretable formulas.
Keywords:
accuracy predictioncomputer visiondeep learningdistance correlationedge AIedge hardwareinference latencyinterpretable machine learningneural architecturesymbolic regression

Related Experiment Videos

  • Developed device-specific inference latency formulas for CPU, GPU, and NPU.

    • Main Results:

    • Batch size and total layer count identified as strongest universal accuracy predictors (dCor: 0.228, 0.174).
    • Learning rate showed the highest MIC (0.226), indicating a non-monotonic accuracy relationship.
    • Symbolic regression models achieved R2 from 0.20 to 0.45 (dataset-specific) and 0.23 (universal), outperforming linear regression.
    • Latency formulas significantly outperformed classical baselines (6.7×-14.8× R2 improvement).

    Conclusions:

    • The proposed framework provides interpretable surrogate models for screening deep vision architectures.
    • Results offer crucial insights for optimizing models under accuracy and latency constraints for edge deployment.
    • Demonstrated the effectiveness of symbolic regression for cross-dataset vision model analysis and prediction.