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 Concept Videos

Natural and Artificial Concepts01:24

Natural and Artificial Concepts

199
In psychology, concepts can be divided into two categories: natural and artificial. Natural concepts are formed through direct or indirect experiences. For example, consider the concept of snow. If you live in a place with regular snowfall, such as Essex Junction, Vermont, you know snow through direct experiences. You’ve seen it fall, touched it, shoveled it, and played in it. You recognize its texture, appearance, and even its smell. In contrast, if you live on an island like Saint...
199
Concepts and Prototypes01:24

Concepts and Prototypes

182
The human nervous system handles vast amounts of information by translating sensory stimuli into neural impulses, which the brain processes, creating thoughts expressed through language or stored as memories. The brain also synthesizes information from emotions and memories, which significantly influence thoughts and behaviors. This intricate process creates a comprehensive mental picture.
The brain organizes this information using concepts, which are mental categories grouping linguistic data,...
182
Neural Circuits01:25

Neural Circuits

1.3K
Neural circuits and neuronal pools are two of the main structures found in the nervous system. Neural circuits are networks of neurons that work together to carry out a specific task or process. They consist of interconnected neurons and glial cells, which provide structural and metabolic support.
Neuronal pools are collections of nerve cells with similar functions and interact through chemical and electrical signals. These pools include both interneurons (the central neural circuit nodes that...
1.3K

You might also read

Related Articles

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

Sort by
Same author

Explainable AI in medicine: challenges of integrating XAI into the future clinical routine.

Frontiers in radiology·2025
Same author

The Importance of Understanding Deep Learning.

Erkenntnis·2024
Same author

ML interpretability: Simple isn't easy.

Studies in history and philosophy of science·2024
Same author

Understanding risk with FOTRES?

AI and ethics·2023
Same author

Machine learning and the quest for objectivity in climate model parameterization.

Climatic change·2023

Related Experiment Video

Updated: Jul 24, 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

583

Methods for identifying emergent concepts in deep neural networks.

Tim Räz1

  • 1University of Bern, Institute of Philosophy, Länggassstrasse 49a, 3012 Bern, Switzerland.

Patterns (New York, N.Y.)
|July 6, 2023
PubMed
Summary
This summary is machine-generated.

Deep neural networks (DNNs) learn concepts, but current detection methods are unreliable due to instance-based concept definition. Combining methods and using synthetic data can improve reliability for understanding DNN concept formation.

Keywords:
TCAVconceptsdeep neural networksfeature visualizationimage classificationinternal representationinterpretabilitynetwork dissection

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

9.2K
Author Spotlight: Addressing Technical and Subjective Challenges in Measuring Classroom Attention
06:37

Author Spotlight: Addressing Technical and Subjective Challenges in Measuring Classroom Attention

Published on: December 15, 2023

3.9K

Related Experiment Videos

Last Updated: Jul 24, 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

583
Deep Neural Networks for Image-Based Dietary Assessment
13:19

Deep Neural Networks for Image-Based Dietary Assessment

Published on: March 13, 2021

9.2K
Author Spotlight: Addressing Technical and Subjective Challenges in Measuring Classroom Attention
06:37

Author Spotlight: Addressing Technical and Subjective Challenges in Measuring Classroom Attention

Published on: December 15, 2023

3.9K

Area of Science:

  • Artificial Intelligence
  • Machine Learning
  • Cognitive Science

Background:

  • Deep neural networks (DNNs) possess internal representations where concepts are learned.
  • Existing methods for detecting concepts in DNNs include network dissection, feature visualization, and concept activation vectors (TCAV).

Purpose of the Study:

  • To critically evaluate current methods for detecting concepts within DNNs.
  • To highlight the limitations of instance-based concept specification and propose solutions.
  • To explore the formation of conceptual spaces in DNNs and the associated trade-offs.

Main Methods:

  • Discussion and critical analysis of existing concept detection techniques (network dissection, feature visualization, TCAV).
  • Exploration of the role of instance selection in concept definition.
  • Consideration of combining methods and using synthetic datasets for improved reliability.
  • Analysis of conceptual spaces shaped by accuracy-compression trade-offs.

Main Results:

  • Current methods provide evidence for DNNs learning non-trivial concept relations.
  • Instance-based concept specification leads to underdetermination and unreliability.
  • Combining methods and using synthetic data can partially mitigate unreliability.
  • Conceptual spaces are crucial for understanding concept formation but lack dedicated study methods.

Conclusions:

  • While DNNs learn concepts, current detection methods have inherent limitations.
  • Improving concept detection reliability requires addressing instance underdetermination.
  • Further research is needed to develop robust methods for studying conceptual spaces in DNNs.