Are Deep Neural Networks Adequate Behavioural Models of Human Visual Perception?

doi:10.1146/annurev-vision-120522-031739

Paper

Are Deep Neural Networks Adequate Behavioural Models of Human Visual Perception?

Published Mar 31, 2023 · Felix Wichmann, Robert Geirhos

Annual review of vision science

Q1 SJR score

16

Citations

0

Influential Citations

Full text Semantic Scholar

Abstract

Deep neural networks (DNNs) are machine learning algorithms that have revolutionized computer vision due to their remarkable successes in tasks like object classification and segmentation. The success of DNNs as computer vision algorithms has led to the suggestion that DNNs may also be good models of human visual perception. In this article, we review evidence regarding current DNNs as adequate behavioral models of human core object recognition. To this end, we argue that it is important to distinguish between statistical tools and computational models and to understand model quality as a multidimensional concept in which clarity about modeling goals is key. Reviewing a large number of psychophysical and computational explorations of core object recognition performance in humans and DNNs, we argue that DNNs are highly valuable scientific tools but that, as of today, DNNs should only be regarded as promising-but not yet adequate-computational models of human core object recognition behavior. On the way, we dispel several myths surrounding DNNs in vision science. Expected final online publication date for the Annual Review of Vision Science, Volume 9 is September 2023. Please see http://www.annualreviews.org/page/journal/pubdates for revised estimates.

Study Snapshot

Key takeawayDeep neural networks are valuable scientific tools but currently only serve as promising computational models of human core object recognition behavior.

PopulationOlder adults (50-71 years)

Sample size24

MethodsObservational

OutcomesBody Mass Index projections

ResultsSocial networks mitigate obesity in older groups.

Sign up to use Study Snapshot

Consensus is limited without an account. Create an account or sign in to get more searches and use the Study Snapshot.

Create an account

Paper

Are Deep Neural Networks Adequate Behavioural Models of Human Visual Perception?

Sign up to use Study Snapshot

References

Factors Affecting Stem Cell-Based Regenerative Approaches in Retinal Degeneration.

Stem cell-based regenerative approaches for retinal degeneration face challenges in cell delivery, recipient immunological microenvironment, and connectivity issues with host circuitry.

Suppressing Retinal Remodeling to Mitigate Vision Loss in Photoreceptor Degenerative Disorders.

Suppressing retinal remodeling in photoreceptor degenerative disorders may help mitigate vision loss by preventing nonhomeostatic changes in retinal neurons.

An Expanding Role for Nonvisual Opsins in Extraocular Light Sensing Physiology.

Nonvisual opsins, such as encephalopsin and neuropsin, play a significant role in extraocular light sensing in mammals, potentially impacting seasonal behavior and camouflage.

Using Natural Scenes to Enhance our Understanding of the Cerebral Cortex's Role in Visual Search.

Using natural scenes for studying the cerebral cortex enhances our understanding of its role in visual search, providing new insights compared to conventional methods.

Impact of Apps as Assistive Devices for Visually Impaired Persons.

Mobile apps for visually impaired individuals have received positive feedback, but rigorous evaluation studies are needed to better understand their impact on daily tasks and quality of life.

Citations

Synthetic And Natural Face Identity Processing Share Common Mechanisms

Deepfake faces and natural faces are perceived and remembered using the same perceptual mechanisms, highlighting the potential risks and benefits of synthetic identities in research and development.

How Aligned are Different Alignment Metrics?

Different alignment metrics for artificial neural networks to human perception are not highly aligned, highlighting the need for integrative benchmarking and fair aggregation of individual metrics.

Multistage maturation optimizes vision

Late development of color vision improves object recognition, highlighting the importance of multistage maturation in the visual system.

Novel intelligent supervised neuro-structures for nonlinear financial crime differential systems

NARX-DLS-based intelligent supervised neuro-structures accurately estimate complex nonlinear financial crime differential systems, aiding in the fight against financial crime.

Solving Bongard Problems With a Visual Language and Pragmatic Constraints

This system, using a symbolic visual vocabulary and Bayesian inference, can effectively infer compositional visual concepts from examples, improving on previous approaches.