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: Jul 4, 2026

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

Deep Neural Networks for Image-Based Dietary Assessment

Published on: March 13, 2021

Layer-specific feature preference in a trained AlexNet model revealed by stylized image analysis.

Nobuhiko Wagatsuma1, Kazuma Ito2, Akinori Hidaka3

  • 1Department of Information Science, Faculty of Science, Toho University, 2-2-1 Miyama, Funabashi, Chiba, 274-8510, Japan. nwagatsuma@is.sci.toho-u.ac.jp.

Scientific Reports
|May 19, 2026
PubMed
Summary

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

Sex-related differences in postoperative complications after pelvic exenteration and their relationship with surgical invasiveness in a multicenter study.

Surgery today·2026
Same author

Circulating Tumor DNA Status and Adjuvant Chemotherapy in Resected Colorectal Liver Metastases.

JAMA oncology·2026
Same author

Prognostic and Predictive Role of Lymph Node Count in Stage II Colon Cancer: A Retrospective Analysis of a Japanese Nationwide Cohort.

Annals of gastroenterological surgery·2026
Same author

Current and Future Perspectives of Adjuvant Therapy for Resected Colorectal Liver Metastases.

Cancers·2026
Same author

Efficacy and safety of second-line fluorouracil, leucovorin, and irinotecan plus ramucirumab according to prior first-line regimens in metastatic colorectal cancer: a multicenter retrospective study.

International journal of clinical oncology·2026
Same author

[Impact of Seasonal Climate Variations on Skin Toxicity Induced by Anti-EGFR Antibody Therapy in Patients with Metastatic Colorectal Cancer].

Gan to kagaku ryoho. Cancer & chemotherapy·2026

Deep convolutional neural networks (DCNNs) show a shift in how they process information. Early layers focus on content and structure, while later layers prioritize style and texture.

Area of Science:

  • Computer Science
  • Artificial Intelligence
  • Neuroscience

Background:

  • Deep convolutional neural networks (DCNNs) excel at object classification.
  • Understanding how DCNNs represent information across layers is crucial but remains unclear.
  • Previous studies noted texture biases at output layers, but internal layer representations are less understood.

Purpose of the Study:

  • To quantitatively investigate the layer-wise evolution of representational preferences in DCNNs.
  • To clarify how internal representations transition from content-based to style-based organization.
  • To provide a benchmark for comparing representation changes in different network architectures.

Main Methods:

  • Utilized the AlexNet architecture for analysis.
Keywords:
AlexNetDeep convolutional neural networkModel neuron preferenceObject classificationUMAP

Related Experiment Videos

Last Updated: Jul 4, 2026

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

Deep Neural Networks for Image-Based Dietary Assessment

Published on: March 13, 2021

  • Employed correlation analyses, Uniform Manifold Approximation and Projection (UMAP), and neighborhood-based metrics.
  • Compared responses of natural and stylized images across network layers.

    • Main Results:

    • Identified a layer-wise transition in AlexNet's representational organization.
    • Early and intermediate convolutional layers prioritize content-related structural cues (edges, contours, layout).
    • Higher fully connected layers exhibit organization based on style-related texture and material properties.

    Conclusions:

    • Demonstrated a quantitative shift from content-to-style encoding across AlexNet layers.
    • Clarified the hierarchical reorganization of representations within DCNNs.
    • Established a foundational understanding for future research on neural network representations.