What is the doggest dog? Examination of typicality perception in ImageNet-trained networks
These videos have been matched automatically. Contact us if they are not relevant.
These videos have been matched automatically. Contact us if they are not relevant.
View abstract on PubMed
Summary
This summary is machine-generated.Deep learning models exhibit human-like prototype perception, identifying typical and atypical examples within categories. This study introduces methods to extract these prototypes and anti-prototypes from 42 diverse neural networks.
Area Of Science
- Cognitive Psychology
- Computer Vision
- Machine Learning
Background
- Prototype Theory explains graded category membership, with prototypes as typical examples and anti-prototypes as atypical ones.
- Understanding prototype perception in AI is crucial for human-like world understanding and accurate concept representation.
- Deep learning models offer diverse architectures suitable for studying cognitive psychology theories like Prototype Theory.
Purpose Of The Study
- To investigate how deep neural networks perceive prototypes and anti-prototypes for basic-level categories.
- To develop and apply methods for extracting and visualizing network-perceived prototypes and anti-prototypes.
- To compare the prototype perception of various deep learning models with human perception.
Main Methods
- Proposed three novel methods to identify prototypes and anti-prototypes within deep networks.
- Developed a visualization technique to assess object centrality within categories as perceived by networks.
- Evaluated 42 diverse network architectures including Convolutional Networks, Vision Transformers, ConvNeXts, and hybrid models trained on ImageNet.
Main Results
- Deep networks demonstrate a significant degree of shared typicality perception across different architectures.
- The prototype perception of these networks largely aligns with human typicality judgments.
- A dataset of per-network prototypes and anti-prototypes was generated for future research.
Conclusions
- Deep learning models exhibit a human-like understanding of category typicality.
- The developed methods enable the study of cognitive phenomena within artificial neural networks.
- Findings facilitate the development of more human-like AI systems and provide insights into model interpretability.
These videos have been matched automatically. Contact us if they are not relevant.
These videos have been matched automatically. Contact us if they are not relevant.
Related Concept Videos
Vision is the result of light being detected and transduced into neural signals by the retina of the eye. This information is then further analyzed and interpreted by the brain. First, light enters the front of the eye and is focused by the cornea and lens onto the retina—a thin sheet of neural tissue lining the back of the eye. Because of refraction through the convex lens of the eye, images are projected onto the retina upside-down and reversed.
Light is absorbed by the rod and cone...
Gestalt principles provide a framework for understanding how humans perceive objects as unified wholes within their context. These principles are essential in explaining the cognitive processes that make sense of complex visual stimuli by organizing them into coherent groups. One fundamental principle is proximity, which posits that objects located close to each other are perceived as a collective group. For instance, when dots are positioned near one another, the visual system interprets them...
Light enters the eye through the cornea, a transparent, dome-shaped surface covering the surface of the eyeball that helps to direct and focus incoming light. This light is then channeled toward the pupil, an adjustable opening whose size is controlled by the iris. The iris, a pigmented muscle, regulates the amount of light entering the eye by contracting or dilating the pupil, thereby ensuring optimal light levels for clear vision.
Once through the pupil, the light passes through the lens, a...